Based on kernel version 4.16.1. Page generated on 2018-04-09 11:53 EST.
1 == Overview == 2 3 Original x86-64 was limited by 4-level paing to 256 TiB of virtual address 4 space and 64 TiB of physical address space. We are already bumping into 5 this limit: some vendors offers servers with 64 TiB of memory today. 6 7 To overcome the limitation upcoming hardware will introduce support for 8 5-level paging. It is a straight-forward extension of the current page 9 table structure adding one more layer of translation. 10 11 It bumps the limits to 128 PiB of virtual address space and 4 PiB of 12 physical address space. This "ought to be enough for anybody" ©. 13 14 QEMU 2.9 and later support 5-level paging. 15 16 Virtual memory layout for 5-level paging is described in 17 Documentation/x86/x86_64/mm.txt 18 19 == Enabling 5-level paging == 20 21 CONFIG_X86_5LEVEL=y enables the feature. 22 23 So far, a kernel compiled with the option enabled will be able to boot 24 only on machines that supports the feature -- see for 'la57' flag in 25 /proc/cpuinfo. 26 27 The plan is to implement boot-time switching between 4- and 5-level paging 28 in the future. 29 30 == User-space and large virtual address space == 31 32 On x86, 5-level paging enables 56-bit userspace virtual address space. 33 Not all user space is ready to handle wide addresses. It's known that 34 at least some JIT compilers use higher bits in pointers to encode their 35 information. It collides with valid pointers with 5-level paging and 36 leads to crashes. 37 38 To mitigate this, we are not going to allocate virtual address space 39 above 47-bit by default. 40 41 But userspace can ask for allocation from full address space by 42 specifying hint address (with or without MAP_FIXED) above 47-bits. 43 44 If hint address set above 47-bit, but MAP_FIXED is not specified, we try 45 to look for unmapped area by specified address. If it's already 46 occupied, we look for unmapped area in *full* address space, rather than 47 from 47-bit window. 48 49 A high hint address would only affect the allocation in question, but not 50 any future mmap()s. 51 52 Specifying high hint address on older kernel or on machine without 5-level 53 paging support is safe. The hint will be ignored and kernel will fall back 54 to allocation from 47-bit address space. 55 56 This approach helps to easily make application's memory allocator aware 57 about large address space without manually tracking allocated virtual 58 address space. 59 60 One important case we need to handle here is interaction with MPX. 61 MPX (without MAWA extension) cannot handle addresses above 47-bit, so we 62 need to make sure that MPX cannot be enabled we already have VMA above 63 the boundary and forbid creating such VMAs once MPX is enabled.