1*ee65728eSMike Rapoport============================== 2*ee65728eSMike Rapoportremap_file_pages() system call 3*ee65728eSMike Rapoport============================== 4*ee65728eSMike Rapoport 5*ee65728eSMike RapoportThe remap_file_pages() system call is used to create a nonlinear mapping, 6*ee65728eSMike Rapoportthat is, a mapping in which the pages of the file are mapped into a 7*ee65728eSMike Rapoportnonsequential order in memory. The advantage of using remap_file_pages() 8*ee65728eSMike Rapoportover using repeated calls to mmap(2) is that the former approach does not 9*ee65728eSMike Rapoportrequire the kernel to create additional VMA (Virtual Memory Area) data 10*ee65728eSMike Rapoportstructures. 11*ee65728eSMike Rapoport 12*ee65728eSMike RapoportSupporting of nonlinear mapping requires significant amount of non-trivial 13*ee65728eSMike Rapoportcode in kernel virtual memory subsystem including hot paths. Also to get 14*ee65728eSMike Rapoportnonlinear mapping work kernel need a way to distinguish normal page table 15*ee65728eSMike Rapoportentries from entries with file offset (pte_file). Kernel reserves flag in 16*ee65728eSMike RapoportPTE for this purpose. PTE flags are scarce resource especially on some CPU 17*ee65728eSMike Rapoportarchitectures. It would be nice to free up the flag for other usage. 18*ee65728eSMike Rapoport 19*ee65728eSMike RapoportFortunately, there are not many users of remap_file_pages() in the wild. 20*ee65728eSMike RapoportIt's only known that one enterprise RDBMS implementation uses the syscall 21*ee65728eSMike Rapoporton 32-bit systems to map files bigger than can linearly fit into 32-bit 22*ee65728eSMike Rapoportvirtual address space. This use-case is not critical anymore since 64-bit 23*ee65728eSMike Rapoportsystems are widely available. 24*ee65728eSMike Rapoport 25*ee65728eSMike RapoportThe syscall is deprecated and replaced it with an emulation now. The 26*ee65728eSMike Rapoportemulation creates new VMAs instead of nonlinear mappings. It's going to 27*ee65728eSMike Rapoportwork slower for rare users of remap_file_pages() but ABI is preserved. 28*ee65728eSMike Rapoport 29*ee65728eSMike RapoportOne side effect of emulation (apart from performance) is that user can hit 30*ee65728eSMike Rapoportvm.max_map_count limit more easily due to additional VMAs. See comment for 31*ee65728eSMike RapoportDEFAULT_MAX_MAP_COUNT for more details on the limit. 32