[lkml]   [2004]   [Sep]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [vm 0/4] replace remap_page_range() with remap_pfn_range()
    On Thu, Sep 23, 2004 at 06:22:22PM -0500, John Fusco wrote:
    >> Everything worked great until we decided that we needed to install 6GB
    >> in this system. The problem is that remap_page_range() uses an unsigned
    >> long as the parameter for a physical address. On IA32, an unsigned long
    >> is 32-bits, but the IA32 is capable of addressing well over 4GB of RAM.
    >> So physical addresses on IA32 must be larger than 32 bits.

    On Thu, Sep 23, 2004 at 07:17:35PM -0700, William Lee Irwin III wrote:
    > Do these patches work for you? Compiletested on sparc64.

    Long-format changelog:

    Resolve physical address overflow issue in the remap_page_range() API
    by replacing it with remap_pfn_range(), which accepts its physical
    address argument as a pfn, hence allowing the use of a single-precision
    physical address argument without the risk of overflow at the API
    boundary. The above issue has hobbled support for various 32-bit
    architectures, including some embedded systems (ppc440 IIRC), caused
    persistent portability issues for sound drivers for legacy systems
    (sparc32; unfortunately this patch alone does not fully resolve those),
    and according to John Fusco's reports, made drivers for some PCI-X
    hardware infeasible to port to recent ia32 PAE enterprise systems. With
    this patch series applied, physical address overflows on 32-bit systems
    caused directly by remap_page_range() are gone forever, and ca. 100LOC
    of cut-and-waste driver code are swept out of existence alongside them.

    -- wli

    P.S.: The existing solution to the sparc32 issue was to pass a double
    precision representation of the physical address as 2 single-
    precision arguments in an API (io_remap_page_range()) whose
    argument corresponding to those two was a single single-
    precision argument on most/all other architectures. The
    sparc32-specific issue requires more work beyond these patches
    to rectify. The most apparent consequence of the API skew is
    that drivers don't compile on sparc32 when they use
    io_remap_page_range() due to passing insufficient arguments,
    or vice-versa for drivers originally written for sparc32.
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2005-03-22 14:06    [W:0.021 / U:1.480 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site