On Sat, Feb 19, 2005 at 12:54:07PM +0900, YAMAMOTO Takashi wrote:
> hi,
> 
> > > > the problem that uvm_km_check_empty() noticed is from the call in
> > > > pmap_init_md().  here we try to claim the virtual space which is already
> > > > backed by PTEs set up back in pmap_bootstrap().  hopefully we can just
> > > > use the normal pmap interfaces to create those mappings later, but I haven't
> > > > tried it yet.
> > > 
> > > is there any reason these areas are reserved using uvm_map,
> > > rather than using the same way as kernel text?  merely to save some va?
> > 
> > no good reason that I know of.  I'd be fine with doing it that way.
> > if you write it, I'll test it.  :-)
> 
> i wrote a patch for mac68k. (attached)

didn't work:

panic: pmap_enter_ptpage: can't get KPT page
Stopped in pid 0.1 () at        netbsd:cpu_Debugger+0x6:        unlk    a6
db> t
cpu_Debugger(2700,c04000,0,345ed8,21662e) + 6
panic(265b53,1002700,346000,1000000,3) + f8
pmap_enter_ptpage(2bda28,1000000) + 24a
pmap_kenter_pa(1000000,346000,3) + 68
uvm_pageboot_alloc(177648,1da00,c830,8006086a,f9000e00) + 7c
uvm_page_init(345f7c,345f80,2ba284,0,64) + c4
uvm_init(25747f,2662fe,1da00,c830,ffffcffc) + 3c


 
> i'm not sure if wasting this kva (thus possibly some static pt pages) is
> acceptable for these platforms, though.

apparently not.  :-)


> if it isn't, i guess it's better to change address layout so that
> the available region is contiguous.  maybe move Sysmap to the end of
> address space?

that sounds fine.

-Chuck