Understanding the Linux Kernel

6.4. Anticipating Linux 2.4

Linux 2.2 has two buddy systems: the first one handles page frames suitable for ISA DMA, while the second one handles page frames not suitable for ISA DMA. Linux 2.4 adds a third buddy system for the high physical memory, that is, for the page frames not permanently mapped by the kernel. Using a high-memory page frame implies changing an entry in a special kernel Page Table to map the page frame physical addresses in the 4 GB linear address space.

Actually, Linux 2.4 views the three portions of RAM as different "zones." Each zone has its own counters and watermarks to monitor the number of free page frames. When a memory allocation request takes place, the kernel first tries to fetch the page frames from the most suitable zone; if it fails, it may fall back on another zone.

The slab allocator is mostly unchanged. However, Linux 2.4 allows a slab allocator cache that is no longer useful to be destroyed. Recall that in Linux 2.2 a slab allocator cache can be dynamically created but not destroyed. Modules that create their own slab allocator cache when loaded are now expected to destroy it when unloaded.