Re: mmap64 for PA

[Matthew Wilcox]

On Thu, Apr 19, 2001 at 09:04:15PM -0700, Ulrich Drepper wrote:
> This kind of change shouldn't be necessary.  You should be able to
> include the PPC version and unconditionally define
> __ASSUME_MMAP2_SYSCALL for PA in kernel-features.h.

I could do that.

> Beside, your code wasn't correct.  Whether any of the lower 12 bits is
> set does not determine whether mmap2 or mmap is used.  If the offset
> is not page aligned this is an error.

Would you like a patch for powerpc/mmap64.c then?

#ifdef __NR_mmap2
  if (
# ifndef __ASSUME_MMAP2_SYSCALL
      ! have_no_mmap2 &&
# endif
      ! (offset & ((1 << PAGE_SHIFT)-1)))
    {

in current CVS.

> Finally, will PA forever have 4k pages?  I thought PA can have
> different page sizes.  What will the mmap2 implementation do in this case?

the kernel hardcodes 12-bits as the `page size' presented to mmap2,
just as sparc32 does.  in fact only PA2.0 can have page sizes other
than 4k, and all pa 2.0 processors can be run in 64-bit mode, so mmap2
becomes irrelevant.