Pager, Reclaim and Working Sets

What is it?

The pager moves memory pages out to disk when RAM runs short and brings them back on the next access. Impossible OS has a basic pagefile swap layer with a clock victim picker and swap-encoded page table entries, but nothing initialises it during a normal boot, so no page is ever swapped outside the unit tests. Working sets, background reclaim and a lazy file-backed pager are not built.

How does it work?

swap.c stores swapped pages in C:\Impossible\System\pagefile.sys in 4 KiB slots. swap_init() reads the slot count from the Registry value HKLM\SYSTEM\Memory\SwapSlots (capped at SWAP_MAX_SLOTS, 256), creates the file and preallocates it. swap_out() writes a page to a free slot and replaces its PTE with a swap-encoded entry (SWAP_ENCODE_PTE() in swap.h); swap_in() reads it back into a fresh frame. The victim is chosen by a clock algorithm over up to CLOCK_MAX_PAGES (512) registered pages (swap_clock_register(), swap_clock_victim()). The slot and clock arrays are global and unlocked.

On a page fault the VMM tries swap_handle_fault() first and mmap_handle_fault() second (vmm.c). Because swap_init() is only called from the test suite, the swap handler declines every fault in a normal boot.

File mappings are eager: mmap.c reads the whole file range with vfs_read() when the mapping is created, because the fault path cannot yet issue blocking file I/O. Mappings live in a window from MMAP_BASE (4 GiB) to MMAP_END (5 GiB), up to MMAP_MAX_REGIONS (64) at a time.

What are its interfaces?

Interface	Purpose
`swap_init()`, `swap_out()`, `swap_in()`, `swap_handle_fault()`	Pagefile slots and swap faults (`swap.h`)
`swap_clock_register()`, `swap_clock_victim()`	Clock replacement candidates
`swap_get_total_slots()`, `swap_get_used_slots()`, `swap_get_free_slots()`	Slot statistics
`mmap()`, `munmap()`, `msync()`, `mmap_handle_fault()`	Eager file and anonymous mappings (`mmap.h`)
`HKLM\SYSTEM\Memory\SwapSlots`	Pagefile size in slots

How do I use it?

There is no operator control yet: no pagefile appears on a booted system, and there are no working-set commands. The swap round trip and the file mapping path are exercised by test_swap.c and test_mmap.c:

bash scripts/test.sh SUITE=mm

What is not implemented yet?

A pagefile manager that starts at boot. Runtime-sized slots, locking, statistics and a degraded-state report instead of silent absence (Pagefile Manager + Swap Metadata Foundation).
Per-process working sets and the rules for which pages may never leave RAM (Working-Set Tracking).
Background reclaim and a modified-page writer (Background Reclaim + Modified-Page Writer).
Lazy file-backed paging with soft and hard faults (Lazy File-Backed Pager).
A replacement policy with refault tracking (Replacement Policy + Refault Tracking).
Working-set APIs and statistics export (Control Surface + Observability Handoff).

How does it compare with Windows 11 and Linux?

Windows 11 pages to pagefile.sys with per-process working sets, trimming, a modified-page writer and demand-paged sections; Linux has swap, kswapd, writeback, a lazy file pager and refault-aware reclaim. Impossible OS has only the slot format and clock picker, unused at runtime, so every row in the roadmap's comparison table is planned. The roadmap's aim is a single pager pipeline that owns residency, reclaim and statistics in one place rather than spread across separate subsystems.