[Qemu-devel] linux-user 32/64 bit mremap() problem

[Qemu-devel] linux-user 32/64 bit mremap() problem

[Vince Weaver]

Hello

This has been discuessed a year ago
   http://www.nabble.com/linux-user-mmap()-for-32-bit-guest-on-64-bit-host-td12934745.html
   http://osdir.com/ml/linux.ports.x86-64.general/2007-10/msg00007.html
but hasn't really been resolved.

The linux-user mremap() syscall implementation will return a 64-bit value 
when running on a 64-bit machine, even when the architecture being 
simulated is only 32-bits.

Has any progress been made on this issue?  I might try implementing the 
workaround the kernel people suggested.

This actually prevents one of the perlbmk spec2k benchmarks from running 
with sparc32plus emulation on an x86_64 machine.

Vince

Re: [Qemu-devel] linux-user 32/64 bit mremap() problem

[Kirill A. Shutemov]

[-- Attachment #1: Type: text/plain, Size: 1014 bytes --]

On Thu, Sep 4, 2008 at 12:01 AM, Vince Weaver <vince@csl.cornell.edu> wrote:

> Hello
>
> This has been discuessed a year ago
>
> http://www.nabble.com/linux-user-mmap()-for-32-bit-guest-on-64-bit-host-td12934745.html<http://www.nabble.com/linux-user-mmap%28%29-for-32-bit-guest-on-64-bit-host-td12934745.html>
>  http://osdir.com/ml/linux.ports.x86-64.general/2007-10/msg00007.html
> but hasn't really been resolved.
>
> The linux-user mremap() syscall implementation will return a 64-bit value
> when running on a 64-bit machine, even when the architecture being simulated
> is only 32-bits.
>
> Has any progress been made on this issue?  I might try implementing the
> workaround the kernel people suggested.
>
> This actually prevents one of the perlbmk spec2k benchmarks from running
> with sparc32plus emulation on an x86_64 machine.
>

I have posted patch to the maillist witch add flag MAP_32BIT to mmap and
mremap calls. It helps on x86_64 host, but other 64-bit host doesn't support
this flag currently.

[-- Attachment #2: Type: text/html, Size: 1505 bytes --]