xref: /linux/Documentation/mm/remap_file_pages.rst (revision daa2be74b1b2302004945b2a5e32424e177cc7da)
1==============================
2remap_file_pages() system call
3==============================
4
5The remap_file_pages() system call is used to create a nonlinear mapping,
6that is, a mapping in which the pages of the file are mapped into a
7nonsequential order in memory. The advantage of using remap_file_pages()
8over using repeated calls to mmap(2) is that the former approach does not
9require the kernel to create additional VMA (Virtual Memory Area) data
10structures.
11
12Supporting of nonlinear mapping requires significant amount of non-trivial
13code in kernel virtual memory subsystem including hot paths. Also to get
14nonlinear mapping work kernel need a way to distinguish normal page table
15entries from entries with file offset (pte_file). Kernel reserves flag in
16PTE for this purpose. PTE flags are scarce resource especially on some CPU
17architectures. It would be nice to free up the flag for other usage.
18
19Fortunately, there are not many users of remap_file_pages() in the wild.
20It's only known that one enterprise RDBMS implementation uses the syscall
21on 32-bit systems to map files bigger than can linearly fit into 32-bit
22virtual address space. This use-case is not critical anymore since 64-bit
23systems are widely available.
24
25The syscall is deprecated and replaced it with an emulation now. The
26emulation creates new VMAs instead of nonlinear mappings. It's going to
27work slower for rare users of remap_file_pages() but ABI is preserved.
28
29One side effect of emulation (apart from performance) is that user can hit
30vm.max_map_count limit more easily due to additional VMAs. See comment for
31DEFAULT_MAX_MAP_COUNT for more details on the limit.
32