> host_map_page (?) uses a MAP_NR, which returns a Bit32u (?) (where is this
> defined, anyway?). It appears that MAP_NR has been replaced by virt_to_page
> in 2.4. However, it (apparently) returns a mem_map_t*, as I gather from
> linux/include/asm-i386/page.h:
>
> #define virt_to_page(kaddr) (mem_map + (__pa(kaddr) >> PAGE_SHIFT))
>
> And mem_map is defined in linux/include/linux/mm.h to be a mem_map_t*
> (which is a *struct). It appears that either some semantics have changed, or
> the original MAP_NR code was flawed, or (most likely) I am missing something
> big here. It appears that host_map_page is supposed to give the host memory
> map for a given guest page, but on that account I am uncertain.
Well, yes, this part of memory management has changed quite a bit in 2.4.
Fortunately, we can ignore the whole issue by just doing everything ourselves.
Try replacing host_map_page with something like this:
Bit32u
host_map_page(void *ptr)
{
Bit32u page = 0;
if (!ptr) return 0;
retrieve_phy_pages(&page, 1, ptr, 4096);
return page;
}
(I've been working on a patch to simplify these issues, but due to lack of
time I haven't finished this yet ...)
> This is the main point at which plex86 does not work for 2.4. With 2.4 coming
> out soon, this had better get fixed. So would somebody more enlighened in
> kernel hacking spread some of that enlightenment around?
Well, I'll try to get it running on 2.4 once I find some time ...
Bye,
Ulrich
P.S. I'll be at Linux Kongress on Thursday and Friday ...
--
Dr. Ulrich Weigand
[EMAIL PROTECTED]
