mm.rst 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157
  1. .. SPDX-License-Identifier: GPL-2.0
  2. =================
  3. Memory Management
  4. =================
  5. Complete virtual memory map with 4-level page tables
  6. ====================================================
  7. .. note::
  8. - Negative addresses such as "-23 TB" are absolute addresses in bytes, counted down
  9. from the top of the 64-bit address space. It's easier to understand the layout
  10. when seen both in absolute addresses and in distance-from-top notation.
  11. For example 0xffffe90000000000 == -23 TB, it's 23 TB lower than the top of the
  12. 64-bit address space (ffffffffffffffff).
  13. Note that as we get closer to the top of the address space, the notation changes
  14. from TB to GB and then MB/KB.
  15. - "16M TB" might look weird at first sight, but it's an easier way to visualize size
  16. notation than "16 EB", which few will recognize at first sight as 16 exabytes.
  17. It also shows it nicely how incredibly large 64-bit address space is.
  18. ::
  19. ========================================================================================================================
  20. Start addr | Offset | End addr | Size | VM area description
  21. ========================================================================================================================
  22. | | | |
  23. 0000000000000000 | 0 | 00007fffffffffff | 128 TB | user-space virtual memory, different per mm
  24. __________________|____________|__________________|_________|___________________________________________________________
  25. | | | |
  26. 0000800000000000 | +128 TB | ffff7fffffffffff | ~16M TB | ... huge, almost 64 bits wide hole of non-canonical
  27. | | | | virtual memory addresses up to the -128 TB
  28. | | | | starting offset of kernel mappings.
  29. __________________|____________|__________________|_________|___________________________________________________________
  30. |
  31. | Kernel-space virtual memory, shared between all processes:
  32. ____________________________________________________________|___________________________________________________________
  33. | | | |
  34. ffff800000000000 | -128 TB | ffff87ffffffffff | 8 TB | ... guard hole, also reserved for hypervisor
  35. ffff880000000000 | -120 TB | ffff887fffffffff | 0.5 TB | LDT remap for PTI
  36. ffff888000000000 | -119.5 TB | ffffc87fffffffff | 64 TB | direct mapping of all physical memory (page_offset_base)
  37. ffffc88000000000 | -55.5 TB | ffffc8ffffffffff | 0.5 TB | ... unused hole
  38. ffffc90000000000 | -55 TB | ffffe8ffffffffff | 32 TB | vmalloc/ioremap space (vmalloc_base)
  39. ffffe90000000000 | -23 TB | ffffe9ffffffffff | 1 TB | ... unused hole
  40. ffffea0000000000 | -22 TB | ffffeaffffffffff | 1 TB | virtual memory map (vmemmap_base)
  41. ffffeb0000000000 | -21 TB | ffffebffffffffff | 1 TB | ... unused hole
  42. ffffec0000000000 | -20 TB | fffffbffffffffff | 16 TB | KASAN shadow memory
  43. __________________|____________|__________________|_________|____________________________________________________________
  44. |
  45. | Identical layout to the 56-bit one from here on:
  46. ____________________________________________________________|____________________________________________________________
  47. | | | |
  48. fffffc0000000000 | -4 TB | fffffdffffffffff | 2 TB | ... unused hole
  49. | | | | vaddr_end for KASLR
  50. fffffe0000000000 | -2 TB | fffffe7fffffffff | 0.5 TB | cpu_entry_area mapping
  51. fffffe8000000000 | -1.5 TB | fffffeffffffffff | 0.5 TB | ... unused hole
  52. ffffff0000000000 | -1 TB | ffffff7fffffffff | 0.5 TB | %esp fixup stacks
  53. ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole
  54. ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space
  55. ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole
  56. ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0
  57. ffffffff80000000 |-2048 MB | | |
  58. ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space
  59. ffffffffff000000 | -16 MB | | |
  60. FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offset
  61. ffffffffff600000 | -10 MB | ffffffffff600fff | 4 kB | legacy vsyscall ABI
  62. ffffffffffe00000 | -2 MB | ffffffffffffffff | 2 MB | ... unused hole
  63. __________________|____________|__________________|_________|___________________________________________________________
  64. Complete virtual memory map with 5-level page tables
  65. ====================================================
  66. .. note::
  67. - With 56-bit addresses, user-space memory gets expanded by a factor of 512x,
  68. from 0.125 PB to 64 PB. All kernel mappings shift down to the -64 PB starting
  69. offset and many of the regions expand to support the much larger physical
  70. memory supported.
  71. ::
  72. ========================================================================================================================
  73. Start addr | Offset | End addr | Size | VM area description
  74. ========================================================================================================================
  75. | | | |
  76. 0000000000000000 | 0 | 00ffffffffffffff | 64 PB | user-space virtual memory, different per mm
  77. __________________|____________|__________________|_________|___________________________________________________________
  78. | | | |
  79. 0100000000000000 | +64 PB | feffffffffffffff | ~16K PB | ... huge, still almost 64 bits wide hole of non-canonical
  80. | | | | virtual memory addresses up to the -64 PB
  81. | | | | starting offset of kernel mappings.
  82. __________________|____________|__________________|_________|___________________________________________________________
  83. |
  84. | Kernel-space virtual memory, shared between all processes:
  85. ____________________________________________________________|___________________________________________________________
  86. | | | |
  87. ff00000000000000 | -64 PB | ff0fffffffffffff | 4 PB | ... guard hole, also reserved for hypervisor
  88. ff10000000000000 | -60 PB | ff10ffffffffffff | 0.25 PB | LDT remap for PTI
  89. ff11000000000000 | -59.75 PB | ff90ffffffffffff | 32 PB | direct mapping of all physical memory (page_offset_base)
  90. ff91000000000000 | -27.75 PB | ff9fffffffffffff | 3.75 PB | ... unused hole
  91. ffa0000000000000 | -24 PB | ffd1ffffffffffff | 12.5 PB | vmalloc/ioremap space (vmalloc_base)
  92. ffd2000000000000 | -11.5 PB | ffd3ffffffffffff | 0.5 PB | ... unused hole
  93. ffd4000000000000 | -11 PB | ffd5ffffffffffff | 0.5 PB | virtual memory map (vmemmap_base)
  94. ffd6000000000000 | -10.5 PB | ffdeffffffffffff | 2.25 PB | ... unused hole
  95. ffdf000000000000 | -8.25 PB | fffffbffffffffff | ~8 PB | KASAN shadow memory
  96. __________________|____________|__________________|_________|____________________________________________________________
  97. |
  98. | Identical layout to the 47-bit one from here on:
  99. ____________________________________________________________|____________________________________________________________
  100. | | | |
  101. fffffc0000000000 | -4 TB | fffffdffffffffff | 2 TB | ... unused hole
  102. | | | | vaddr_end for KASLR
  103. fffffe0000000000 | -2 TB | fffffe7fffffffff | 0.5 TB | cpu_entry_area mapping
  104. fffffe8000000000 | -1.5 TB | fffffeffffffffff | 0.5 TB | ... unused hole
  105. ffffff0000000000 | -1 TB | ffffff7fffffffff | 0.5 TB | %esp fixup stacks
  106. ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole
  107. ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space
  108. ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole
  109. ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0
  110. ffffffff80000000 |-2048 MB | | |
  111. ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space
  112. ffffffffff000000 | -16 MB | | |
  113. FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offset
  114. ffffffffff600000 | -10 MB | ffffffffff600fff | 4 kB | legacy vsyscall ABI
  115. ffffffffffe00000 | -2 MB | ffffffffffffffff | 2 MB | ... unused hole
  116. __________________|____________|__________________|_________|___________________________________________________________
  117. Architecture defines a 64-bit virtual address. Implementations can support
  118. less. Currently supported are 48- and 57-bit virtual addresses. Bits 63
  119. through to the most-significant implemented bit are sign extended.
  120. This causes hole between user space and kernel addresses if you interpret them
  121. as unsigned.
  122. The direct mapping covers all memory in the system up to the highest
  123. memory address (this means in some cases it can also include PCI memory
  124. holes).
  125. We map EFI runtime services in the 'efi_pgd' PGD in a 64Gb large virtual
  126. memory window (this size is arbitrary, it can be raised later if needed).
  127. The mappings are not part of any other kernel PGD and are only available
  128. during EFI runtime calls.
  129. Note that if CONFIG_RANDOMIZE_MEMORY is enabled, the direct mapping of all
  130. physical memory, vmalloc/ioremap space and virtual memory map are randomized.
  131. Their order is preserved but their base will be offset early at boot time.
  132. Be very careful vs. KASLR when changing anything here. The KASLR address
  133. range must not overlap with anything except the KASAN shadow area, which is
  134. correct as KASAN disables KASLR.
  135. For both 4- and 5-level layouts, the STACKLEAK_POISON value in the last 2MB
  136. hole: ffffffffffff4111