Re: Status Bits and Super Pages

[Neal H. Walfield]

At Thu, 23 Aug 2007 16:05:30 +0100,
Espen Skoglund wrote:
> The 64MB mapping will effectively be split into page sizes supported
> by the kernel/hardware.  On ia32 this will result in 16 * 4MB pages
> mapped to the memory manager.  Each of these pages will have status
> bits associated with them that you can query individually.
> Additionally you can query the bits of pages mapped from the resource
> manager *into* other address spaces (without taking its current page
> into account).  This also works if you split a mapping into smaller
> page sized, i.e., you can query individual 4KB pages that you have
> mapped out of a 4MB page in your own address space.
> 
> Note that you'll have to enable the "new" mapping databse for this to
> work.

Excellent.  This is exactly what I was looking for.

One clarification: if the manager has the large mappings from sigma0,
the clients have small mappings from the manager, and the manager
accesses a page that is mapped to a client (e.g., writes to it), the
dirty status bit will only be returned when querying the larger page,
not the smaller page.  However, this is not true if both have small
page mappings.

Thanks,
Neal