memory.rst

  1=================================
  2Kernel Memory Layout on ARM Linux
  3=================================
  4
  5		Russell King <rmk@arm.linux.org.uk>
  6
  7		     November 17, 2005 (2.6.15)
  8
  9This document describes the virtual memory layout which the Linux
 10kernel uses for ARM processors.  It indicates which regions are
 11free for platforms to use, and which are used by generic code.
 12
 13The ARM CPU is capable of addressing a maximum of 4GB virtual memory
 14space, and this must be shared between user space processes, the
 15kernel, and hardware devices.
 16
 17As the ARM architecture matures, it becomes necessary to reserve
 18certain regions of VM space for use for new facilities; therefore
 19this document may reserve more VM space over time.
 20
 21=============== =============== ===============================================
 22Start		End		Use
 23=============== =============== ===============================================
 24ffff8000	ffffffff	copy_user_page / clear_user_page use.
 25				For SA11xx and Xscale, this is used to
 26				setup a minicache mapping.
 27
 28ffff4000	ffffffff	cache aliasing on ARMv6 and later CPUs.
 29
 30ffff1000	ffff7fff	Reserved.
 31				Platforms must not use this address range.
 32
 33ffff0000	ffff0fff	CPU vector page.
 34				The CPU vectors are mapped here if the
 35				CPU supports vector relocation (control
 36				register V bit.)
 37
 38fffe0000	fffeffff	XScale cache flush area.  This is used
 39				in proc-xscale.S to flush the whole data
 40				cache. (XScale does not have TCM.)
 41
 42fffe8000	fffeffff	DTCM mapping area for platforms with
 43				DTCM mounted inside the CPU.
 44
 45fffe0000	fffe7fff	ITCM mapping area for platforms with
 46				ITCM mounted inside the CPU.
 47
 48ffc80000	ffefffff	Fixmap mapping region.  Addresses provided
 49				by fix_to_virt() will be located here.
 50
 51ffc00000	ffc7ffff	Guard region
 52
 53ff800000	ffbfffff	Permanent, fixed read-only mapping of the
 54				firmware provided DT blob
 55
 56fee00000	feffffff	Mapping of PCI I/O space. This is a static
 57				mapping within the vmalloc space.
 58
 59VMALLOC_START	VMALLOC_END-1	vmalloc() / ioremap() space.
 60				Memory returned by vmalloc/ioremap will
 61				be dynamically placed in this region.
 62				Machine specific static mappings are also
 63				located here through iotable_init().
 64				VMALLOC_START is based upon the value
 65				of the high_memory variable, and VMALLOC_END
 66				is equal to 0xff800000.
 67
 68PAGE_OFFSET	high_memory-1	Kernel direct-mapped RAM region.
 69				This maps the platforms RAM, and typically
 70				maps all platform RAM in a 1:1 relationship.
 71
 72PKMAP_BASE	PAGE_OFFSET-1	Permanent kernel mappings
 73				One way of mapping HIGHMEM pages into kernel
 74				space.
 75
 76MODULES_VADDR	MODULES_END-1	Kernel module space
 77				Kernel modules inserted via insmod are
 78				placed here using dynamic mappings.
 79
 80TASK_SIZE	MODULES_VADDR-1	KASAn shadow memory when KASan is in use.
 81				The range from MODULES_VADDR to the top
 82				of the memory is shadowed here with 1 bit
 83				per byte of memory.
 84
 8500001000	TASK_SIZE-1	User space mappings
 86				Per-thread mappings are placed here via
 87				the mmap() system call.
 88
 8900000000	00000fff	CPU vector page / null pointer trap
 90				CPUs which do not support vector remapping
 91				place their vector page here.  NULL pointer
 92				dereferences by both the kernel and user
 93				space are also caught via this mapping.
 94=============== =============== ===============================================
 95
 96Please note that mappings which collide with the above areas may result
 97in a non-bootable kernel, or may cause the kernel to (eventually) panic
 98at run time.
 99
100Since future CPUs may impact the kernel mapping layout, user programs
101must not access any memory which is not mapped inside their 0x0001000
102to TASK_SIZE address range.  If they wish to access these areas, they
103must set up their own mappings using open() and mmap().