kswapd

kswapd is a memory management kernel thread in the Linux kernel that handles page reclamation to free memory when the system runs low on free pages. It traditionally runs per NUMA node, with one kswapd instance managing reclaim for that node's memory zones. Its primary role is to keep a healthy pool of free pages by evicting or reclaiming pages from various memory pools, including page cache and anonymous memory.

The reclamation process is triggered when free memory falls below predefined watermarks. When activated, kswapd works in the background to reclaim memory, attempting to free pages by moving them off the LRU lists and, if necessary, swapping them out to disk. It interacts with the page allocator and the swap subsystem and can target both anonymous pages and file-backed caches. The goal is to satisfy future allocation requests without blocking or triggering an out-of-memory condition.

kswapd uses the kernel’s shrinker framework to reclaim memory, iterating over per-zone LRU lists and coordinating with swap and the page cache. If background reclaim cannot keep up with pressure, the system may enter a higher reclaim mode or eventually trigger the OOM killer as a last resort. In some scenarios, processes may perform direct reclaim themselves during page faults if kswapd is not keeping pace.

The behavior of kswapd is influenced by memory pressure and tunables such as swappiness, which affects how aggressively the system prefers swapping out pages. High swap pressure or limited swap space can increase kswapd activity and impact allocation latency. Monitoring tools like vmstat and procfs entries provide visibility into reclaim activity and memory pressure.