Results 1 to 8 of 8

Thread: Linux support for 1tb ram on ARM

  1. #1
    Join Date
    Feb 2008
    Location
    Linuxland
    Posts
    5,187

    Default Linux support for 1tb ram on ARM

    The cortex-A15 cpus support 1tb of ram (previous ones top at 4gb). Linux patches at:

    https://lkml.org/lkml/2011/5/8/37

  2. #2
    Join Date
    Jul 2008
    Location
    Berlin, Germany
    Posts
    835

    Default

    Nothing to get too excited about. LPAE is not designed to give an OS instance access to more than 32 bit address space. So you can have several VMs with 4 GB each at most.

  3. #3
    Join Date
    Jan 2008
    Posts
    772

    Default

    Quote Originally Posted by chithanh View Post
    Nothing to get too excited about. LPAE is not designed to give an OS instance access to more than 32 bit address space. So you can have several VMs with 4 GB each at most.
    There might be some VPS providers out there interested in just such a configuration.

  4. #4
    Join Date
    Jul 2008
    Location
    Berlin, Germany
    Posts
    835

    Default

    I think that it is more likely intended for mobile devices, where the entity who has control over the device runs the user interface/media/DRM/etc. functions in one VM, while the owner can run his own applications in a separate sandboxed VM.

  5. #5
    Join Date
    Feb 2008
    Location
    Linuxland
    Posts
    5,187

    Default

    This is the first step towards proper ARM servers, and having linux support ready before the cpus ship is nice.

  6. #6
    Join Date
    Sep 2006
    Posts
    213

    Default

    Quote Originally Posted by curaga View Post
    This is the first step towards proper ARM servers
    Don't kid yourself. PAE is a horrible crutch on x86 and will be the same on ARM.
    ARM64 (like MIPS64) with a properly though-out 32-bit compat layer (for all existing
    applications) would have been a much better idea, but I guess they needed a quick
    solution on top of the existing arm ecosystem to keep their biggest licensees happy.

  7. #7
    Join Date
    May 2011
    Posts
    353

    Default

    This would mean that the ARM architecture is split between ARM32 and ARM64, because ARM's that don't support 64-bit then fail to run Linux. But anyway, when NVidia's Project Denver ARM64 CPU comes out somewhere around 2013, I am sure that there will be someone starting ARM64 development in the kernel.

  8. #8
    Join Date
    Sep 2006
    Posts
    213

    Default

    Quote Originally Posted by AlbertP View Post
    This would mean that the ARM architecture is split between ARM32 and ARM64, because ARM's that don't support 64-bit then fail to run Linux. But anyway, when NVidia's Project Denver ARM64 CPU comes out somewhere around 2013, I am sure that there will be someone starting ARM64 development in the kernel.
    It's still 32bit instruction set, just the address space the MMU can handle has been
    extended. This is not uncommon (I have a MIPS32 system which has a 36bit address space, 32bit PPC can do the same I believe), but applications are still limited to their theoretical 4GB, which is a bit useless for larger server-type stuff (think working with large matrices, caches, ...).
    I'm pretty sure prototype code to support linux on ARM64 exists. After all, if you have
    the HDL to describe the CPU, you can simulate it or run it on a slow FPGA.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •