On Sat, Apr 06, 2002 at 06:16:40PM +0900, Toru Nishimura wrote:

 > There is no siginificant rationale to have PTE_BASE below KERNEL_BASE,
 > which looks inherited from acent i386 arrangement.  The address range of
 > PTE_BASE and APTE_BASE is considered subject to process context
 > switch even if it was maintained in kernel space and disjoint to user address
 > space.   It'd be better to move adjacent to APTE_BASE.  sh3 folks, it's not
 > optimal to have and maintain the pmap as a copy of i386.  It could be
 > implemented in a saner (clean, handy and managable) way.

Actually, by keeping it below KERNEL_BASE, it allows you to enforce an
invariant: that L1 descriptors that map >= KERNEL_BASE are the same for
each process.

With PTE_BASE the way it is, you automatically get the new processes's
PTEs when you switch the translation table base.  It's quite convenient.

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>