Subject: Re: Kernel copyin/out optimizations for ARM...
To: None <Richard.Earnshaw@arm.com>
From: Jason R Thorpe <thorpej@wasabisystems.com>
List: port-arm
Date: 03/14/2002 08:14:35
On Thu, Mar 14, 2002 at 11:26:29AM +0000, Richard Earnshaw wrote:

 > On reflection, maybe not.  IIRC the page tables for the current process 
 > are mapped linearly at PROCESS_PAGE_TBLS_BASE.  So they are there in an 
 > single array.  What happens if a physical page is missing from that map 
 > and we try to dereference it, I've no idea.  I hope we won't end up 
 > mapping one in, just so that we can decide that it doesn't contain a valid 
 > mapping.

The pmap is supposed to check that the L1 slot for a VA is valid before
touching the L2 table.  If you reference a nonexistent L2 table, you'll
get a fatal data abort (i.e. the VM fault handler won't find a vm_map
mapping for that space and will panic), which is what you want.

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