1                Kernel Memory Layout on ARM Linux
   3                Russell King <>
   4                     November 17, 2005 (2.6.15)
   6This document describes the virtual memory layout which the Linux
   7kernel uses for ARM processors.  It indicates which regions are
   8free for platforms to use, and which are used by generic code.
  10The ARM CPU is capable of addressing a maximum of 4GB virtual memory
  11space, and this must be shared between user space processes, the
  12kernel, and hardware devices.
  14As the ARM architecture matures, it becomes necessary to reserve
  15certain regions of VM space for use for new facilities; therefore
  16this document may reserve more VM space over time.
  18Start           End             Use
  20ffff8000        ffffffff        copy_user_page / clear_user_page use.
  21                                For SA11xx and Xscale, this is used to
  22                                setup a minicache mapping.
  24ffff4000        ffffffff        cache aliasing on ARMv6 and later CPUs.
  26ffff1000        ffff7fff        Reserved.
  27                                Platforms must not use this address range.
  29ffff0000        ffff0fff        CPU vector page.
  30                                The CPU vectors are mapped here if the
  31                                CPU supports vector relocation (control
  32                                register V bit.)
  34fffe0000        fffeffff        XScale cache flush area.  This is used
  35                                in proc-xscale.S to flush the whole data
  36                                cache. (XScale does not have TCM.)
  38fffe8000        fffeffff        DTCM mapping area for platforms with
  39                                DTCM mounted inside the CPU.
  41fffe0000        fffe7fff        ITCM mapping area for platforms with
  42                                ITCM mounted inside the CPU.
  44fff00000        fffdffff        Fixmap mapping region.  Addresses provided
  45                                by fix_to_virt() will be located here.
  47ffc00000        ffefffff        DMA memory mapping region.  Memory returned
  48                                by the dma_alloc_xxx functions will be
  49                                dynamically mapped here.
  51ff000000        ffbfffff        Reserved for future expansion of DMA
  52                                mapping region.
  54VMALLOC_START   VMALLOC_END-1   vmalloc() / ioremap() space.
  55                                Memory returned by vmalloc/ioremap will
  56                                be dynamically placed in this region.
  57                                Machine specific static mappings are also
  58                                located here through iotable_init().
  59                                VMALLOC_START is based upon the value
  60                                of the high_memory variable, and VMALLOC_END
  61                                is equal to 0xff000000.
  63PAGE_OFFSET     high_memory-1   Kernel direct-mapped RAM region.
  64                                This maps the platforms RAM, and typically
  65                                maps all platform RAM in a 1:1 relationship.
  67PKMAP_BASE      PAGE_OFFSET-1   Permanent kernel mappings
  68                                One way of mapping HIGHMEM pages into kernel
  69                                space.
  71MODULES_VADDR   MODULES_END-1   Kernel module space
  72                                Kernel modules inserted via insmod are
  73                                placed here using dynamic mappings.
  7500001000        TASK_SIZE-1     User space mappings
  76                                Per-thread mappings are placed here via
  77                                the mmap() system call.
  7900000000        00000fff        CPU vector page / null pointer trap
  80                                CPUs which do not support vector remapping
  81                                place their vector page here.  NULL pointer
  82                                dereferences by both the kernel and user
  83                                space are also caught via this mapping.
  85Please note that mappings which collide with the above areas may result
  86in a non-bootable kernel, or may cause the kernel to (eventually) panic
  87at run time.
  89Since future CPUs may impact the kernel mapping layout, user programs
  90must not access any memory which is not mapped inside their 0x0001000
  91to TASK_SIZE address range.  If they wish to access these areas, they
  92must set up their own mappings using open() and mmap().
  93 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.