GFP_ATOMIC is a flagspec (Get Free Pages atomic) that changes how the allocator behaves. In the labyrinth, think of it as a sprinting path where you cannot stop (sleep) under any circumstances.
// In a network driver's interrupt handler (atomic context) struct page *excl_page; gfp_t flags = GFP_ATOMIC | __GFP_ZERO;excl_page = labyrinth_alloc_page_gfp_atomic_exclusive(flags, 0); if (!excl_page) // Emergency path: even atomic reserves are exhausted goto drop_packet;
// Because 'exclusive' is set, we can skip certain locking set_page_private(excl_page, MY_MAGIC);define labyrinth void allocpagegfpatomic exclusive
The keyword define labyrinth void allocpagegfpatomic exclusive is not a valid C statement, but it is a powerful specification. It defines a set of constraints: GFP_ATOMIC is a flagspec (Get Free Pages atomic)
alloc_page() (and its relatives __get_free_page(), alloc_pages()) is the function that hands you a key to a physical page frame (usually 4 KB on common architectures). Unlike kmalloc() which gives arbitrary-sized byte chunks, alloc_page() works at the page granularity — the fundamental unit of memory mapping.
Typical usage:
struct page *page = alloc_page(GFP_KERNEL);
if (!page)
return -ENOMEM; // The labyrinth has no free rooms
void *vaddr = page_address(page);
When to use: Device drivers that need DMA buffers, page tables themselves, or any scenario where you need to control physical page properties (e.g., contiguous memory, cache flushing).