Wine and mmap

[Brian Fundakowski Feldman]

On Sat, Aug 28, 2004 at 11:36:12AM +1000, John Birrell wrote:
> On Fri, Aug 27, 2004 at 07:05:02PM -0400, Brian Fundakowski Feldman wrote:
> > Can you be certain that 100% of kernel changes to support WINE are present
> > after fixing mmap?  I heard WINE was still broken, so I haven't bothered
> > trying.
> 
> I think that 100% of the kernel memory changes required to support
> Wine are related to mmap. Wine successfully maps the address space it
> wants before dlopen'ing shared libraries. It's the kernel's choice of
> map address that fails.
> 
> There is at least one problem with Wine using libpthread. Whether that relates
> to kernel or user-space code remains to be seen. I think the thread problem
> is a separate issue.
> 
> FWIW, I see no hurry in changing the mmap address allocation in CVS. This
> issue has been around for months and *very* few people have commented. The
> other BSD's are in the same boat.
> 
> I intend to test the patch written by Anish and move on to the next Wine
> problem 'when I get a chance' (TM). When Wine works again, albeit with a few
> patches, then it will be time to find an implementation that people find
> palatable. I thought that raising the issue on -arch like I did would have
> at least got some response (even a single reply). Linux chooses random mmap
> addresses for good reason. It would seem sensible to do that in FreeBSD too
> (for the same reason).

Is the problem that it specifically runs out of memory area but it's not
checking where it needs to?  If so, I think you could hash out the solution
with a wrapper/workaround mmap() in userland first.

-- 
Brian Fundakowski Feldman                           \'[ FreeBSD ]''''''''''\
  <> green at FreeBSD.org                               \  The Power to Serve! \
 Opinions expressed are my own.                       \,,,,,,,,,,,,,,,,,,,,,,\