feat(mm): migrate to PageAlloc, FrameAlloc #2002

mkroening · 2025-10-21T14:14:50Z

Similar to #1994, this makes the internals of the physical and virtual memory allocation private. Note that this will slow down parts of paging right now, since it is not possible to hold a global FrameAlloc lock anymore. This means the lock has to be reacquired again and again until we improve the paging code or the FrameAlloc code.

Depends on #2008.
Closes #1994.

github-actions

Benchmark Results

Benchmark	Current: `786d4e9`	Previous: `80d7cc6`	Performance Ratio
startup_benchmark Build Time	`109.21` s	`111.54` s	`0.98`
startup_benchmark File Size	`0.91` MB	`0.89` MB	`1.01`
Startup Time - 1 core	`0.91` s (`±0.03` s)	`0.90` s (`±0.03` s)	`1.02`
Startup Time - 2 cores	`0.91` s (`±0.03` s)	`0.90` s (`±0.04` s)	`1.01`
Startup Time - 4 cores	`0.92` s (`±0.03` s)	`0.90` s (`±0.04` s)	`1.02`
multithreaded_benchmark Build Time	`111.64` s	`110.94` s	`1.01`
multithreaded_benchmark File Size	`1.01` MB	`1.00` MB	`1.01`
Multithreaded Pi Efficiency - 2 Threads	`90.87` % (`±9.00` %)	`86.63` % (`±8.31` %)	`1.05`
Multithreaded Pi Efficiency - 4 Threads	`44.15` % (`±4.21` %)	`43.85` % (`±3.75` %)	`1.01`
Multithreaded Pi Efficiency - 8 Threads	`25.47` % (`±2.03` %)	`24.82` % (`±1.93` %)	`1.03`
micro_benchmarks Build Time	`255.58` s	`261.05` s	`0.98`
micro_benchmarks File Size	`1.01` MB	`1.01` MB	`1.01`
Scheduling time - 1 thread	`135.35` ticks (`±37.35` ticks)	`130.02` ticks (`±29.33` ticks)	`1.04`
Scheduling time - 2 threads	`77.00` ticks (`±17.59` ticks)	`74.47` ticks (`±19.53` ticks)	`1.03`
Micro - Time for syscall (getpid)	`8.31` ticks (`±4.31` ticks)	`6.69` ticks (`±3.39` ticks)	`1.24`
Memcpy speed - (built_in) block size 4096	`59053.09` MByte/s (`±42425.30` MByte/s)	`61589.06` MByte/s (`±43639.11` MByte/s)	`0.96`
Memcpy speed - (built_in) block size 1048576	`23561.68` MByte/s (`±21470.34` MByte/s)	`22767.71` MByte/s (`±20609.88` MByte/s)	`1.03`
Memcpy speed - (built_in) block size 16777216	`15328.77` MByte/s (`±12685.30` MByte/s)	`14645.26` MByte/s (`±12130.00` MByte/s)	`1.05`
Memset speed - (built_in) block size 4096	`59528.38` MByte/s (`±42735.96` MByte/s)	`61902.61` MByte/s (`±43834.96` MByte/s)	`0.96`
Memset speed - (built_in) block size 1048576	`24018.20` MByte/s (`±21662.29` MByte/s)	`23121.96` MByte/s (`±20732.33` MByte/s)	`1.04`
Memset speed - (built_in) block size 16777216	`15798.83` MByte/s (`±13006.40` MByte/s)	`15049.49` MByte/s (`±12384.98` MByte/s)	`1.05`
Memcpy speed - (rust) block size 4096	`55765.23` MByte/s (`±41139.86` MByte/s)	`51062.02` MByte/s (`±37732.03` MByte/s)	`1.09`
Memcpy speed - (rust) block size 1048576	`23636.86` MByte/s (`±21020.91` MByte/s)	`21034.25` MByte/s (`±18494.59` MByte/s)	`1.12`
Memcpy speed - (rust) block size 16777216	`15516.08` MByte/s (`±13081.15` MByte/s)	`13683.17` MByte/s (`±11267.50` MByte/s)	`1.13`
Memset speed - (rust) block size 4096	`56038.64` MByte/s (`±41276.95` MByte/s)	`52107.73` MByte/s (`±38544.76` MByte/s)	`1.08`
Memset speed - (rust) block size 1048576	`24002.94` MByte/s (`±21148.69` MByte/s)	`21343.88` MByte/s (`±18636.73` MByte/s)	`1.12`
Memset speed - (rust) block size 16777216	`15838.53` MByte/s (`±13244.53` MByte/s)	`13980.79` MByte/s (`±11428.92` MByte/s)	`1.13`
alloc_benchmarks Build Time	`238.77` s	`238.48` s	`1.00`
alloc_benchmarks File Size	`0.97` MB	`0.96` MB	`1.01`
Allocations - Allocation success	`100.00` %	`100.00` %	`1`
Allocations - Deallocation success	`100.00` %	`100.00` %	`1`
Allocations - Pre-fail Allocations	`100.00` %	`100.00` %	`1`
Allocations - Average Allocation time	`15990.58` Ticks (`±735.75` Ticks)	`15620.57` Ticks (`±710.75` Ticks)	`1.02`
Allocations - Average Allocation time (no fail)	`15990.58` Ticks (`±735.75` Ticks)	`15620.57` Ticks (`±710.75` Ticks)	`1.02`
Allocations - Average Deallocation time	`1974.50` Ticks (`±791.12` Ticks)	`2001.83` Ticks (`±1012.97` Ticks)	`0.99`
mutex_benchmark Build Time	`243.75` s	`243.50` s	`1.00`
mutex_benchmark File Size	`1.02` MB	`1.01` MB	`1.01`
Mutex Stress Test Average Time per Iteration - 1 Threads	`26.20` ns (`±6.14` ns)	`25.56` ns (`±5.79` ns)	`1.03`
Mutex Stress Test Average Time per Iteration - 2 Threads	`24.66` ns (`±3.27` ns)	`24.64` ns (`±3.12` ns)	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

mkroening · 2025-10-22T14:28:30Z

@m-mueller678, what do you think about PageAlloc and FrameAlloc instead of going back to free-standing functions? Feel free to review. :)

m-mueller678

Looks good to me, safe for the few split unsafe blocks. I like the change from free functions to methods.

m-mueller678 · 2025-10-22T14:46:55Z

src/arch/riscv64/mm/mod.rs

 pub fn init() {
-	crate::mm::physicalmem::init();
-	crate::mm::virtualmem::init();
+	unsafe {


could be merged into a single unsafe block

Having these operations in separate blocks was on purpose. See clippy::multiple_unsafe_ops_per_block. Arguably, the situation is far from ideal in the kernel still, but nevertheless, I think this is a step in the right direction.

This function itself should be unsafe too, I just noticed. Let's change that.

src/arch/aarch64/mm/mod.rs

src/arch/x86_64/mm/mod.rs

m-mueller678 · 2025-10-22T14:54:32Z

src/mm/page_range_alloc.rs

+		Ok(Self(range, PhantomData))
+	}
+
+	pub unsafe fn from_raw(range: PageRange) -> Self {


Feels like there should be an into_raw too for symmetry.

Makes sense. I added it. 👍

m-mueller678 · 2025-10-22T14:54:35Z

src/mm/page_range_alloc.rs

+	unsafe fn deallocate(range: PageRange);
+}
+
+pub struct PageRangeBox<A: PageRangeAllocator>(PageRange, PhantomData<A>);


I very much like the idea of PageRangeBox 👍

m-mueller678 · 2025-10-22T14:57:16Z

src/mm/physicalmem.rs

 	InterruptTicketMutex::new(FreeList::new());
 pub static TOTAL_MEMORY: AtomicUsize = AtomicUsize::new(0);

+pub struct FrameAlloc;


I like the idea of having these as methods rather than free standing functions. Both to enable PageRangeBox and to have something natural to implement x86_64's FrameAllocator on.

Co-authored-by: m-mueller678 <[email protected]>

This replaces uses of KERNEL_FREE_LIST with PageAlloc. This replaces uses of PHYSICAL_FREE_LIST with FrameAlloc. Note that this will slow down parts of x86-64 paging right now, since it is not possible to hold a global `FrameAlloc` lock anymore. This means the lock has to be reacquired again and again until we improve the paging code or the `FrameAlloc` code. Co-authored-by: m-mueller678 <[email protected]>

mkroening · 2025-10-23T15:09:13Z

With #2008 merged, this should now be ready too.

mkroening self-assigned this Oct 21, 2025

mkroening force-pushed the page_alloc branch 4 times, most recently from 374c53b to d32f419 Compare October 21, 2025 15:11

github-actions bot reviewed Oct 22, 2025

View reviewed changes

mkroening force-pushed the page_alloc branch from e3a2fde to 6cbf2e3 Compare October 22, 2025 14:15

mkroening mentioned this pull request Oct 22, 2025

fix(mman): don't deallocate PROT_NONE frames #2008

Merged

m-mueller678 suggested changes Oct 22, 2025

View reviewed changes

mkroening and others added 3 commits October 23, 2025 16:46

feat(mm): add PageAlloc, FrameAlloc

e72afef

Co-authored-by: m-mueller678 <[email protected]>

feat(mm): add PageRangeBox, PageBox, FrameBox

0887b57

mkroening force-pushed the page_alloc branch from c771230 to e72afef Compare October 23, 2025 15:07

mkroening marked this pull request as ready for review October 23, 2025 15:09

refactor(x86_64): migrate to PageBox

786d4e9

mkroening force-pushed the page_alloc branch from 22cdfbb to 786d4e9 Compare October 24, 2025 14:17

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(mm): migrate to PageAlloc, FrameAlloc #2002

feat(mm): migrate to PageAlloc, FrameAlloc #2002

Uh oh!

mkroening commented Oct 21, 2025 •

edited

Loading

Uh oh!

github-actions bot left a comment •

edited

Loading

Uh oh!

mkroening commented Oct 22, 2025

Uh oh!

m-mueller678 left a comment

Uh oh!

m-mueller678 Oct 22, 2025 •

edited

Loading

Uh oh!

mkroening Oct 23, 2025

Uh oh!

Uh oh!

Uh oh!

m-mueller678 Oct 22, 2025

Uh oh!

mkroening Oct 23, 2025

Uh oh!

m-mueller678 Oct 22, 2025

Uh oh!

m-mueller678 Oct 22, 2025

Uh oh!

mkroening commented Oct 23, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

feat(mm): migrate to PageAlloc, FrameAlloc #2002

Are you sure you want to change the base?

feat(mm): migrate to PageAlloc, FrameAlloc #2002

Uh oh!

Conversation

mkroening commented Oct 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Benchmark Results

Uh oh!

mkroening commented Oct 22, 2025

Uh oh!

m-mueller678 left a comment

Choose a reason for hiding this comment

Uh oh!

m-mueller678 Oct 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mkroening Oct 23, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

m-mueller678 Oct 22, 2025

Choose a reason for hiding this comment

Uh oh!

mkroening Oct 23, 2025

Choose a reason for hiding this comment

Uh oh!

m-mueller678 Oct 22, 2025

Choose a reason for hiding this comment

Uh oh!

m-mueller678 Oct 22, 2025

Choose a reason for hiding this comment

Uh oh!

mkroening commented Oct 23, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

mkroening commented Oct 21, 2025 •

edited

Loading

github-actions bot left a comment •

edited

Loading

m-mueller678 Oct 22, 2025 •

edited

Loading