A Friendly Tour of Process Memory on Linux

6 months ago

Linux creates an illusion of continuous memory for processes by managing memory in pages, not as a single block.
Process memory is managed through virtual memory areas (VMAs), which are contiguous ranges with specific permissions and backing sources.
Memory mappings can be changed using `mmap`, `mprotect`, and `munmap`, with actual memory allocation happening on first access (lazy allocation).
Page faults occur when accessing unmapped memory, triggering the kernel to allocate or load the required pages.
Copy-on-write (CoW) is used during `fork()` and with `MAP_PRIVATE` mappings to share memory until modifications are made.
Transparent Huge Pages (THP) improve performance by using larger pages (e.g., 2 MiB) to reduce TLB pressure.
The kernel defends against Meltdown using Page Table Isolation (PTI), which separates user and kernel page tables.
Tools like `/proc/<pid>/maps` and `/proc/<pid>/smaps` provide insights into process memory usage and mappings.
Memory permissions (e.g., W^X) and TLB invalidations can introduce small delays, especially during `mprotect` calls.
Common issues include misaligned mappings (`EINVAL`), out-of-memory errors (`ENOMEM`), and `SIGBUS` when accessing beyond file boundaries.

Hasty Briefsbeta