Re: Nick's core remove PageReserved broke vmware...

[Nick Piggin <nickpiggin@yahoo.com.au>]

Petr Vandrovec wrote:
> Hello Nick,
>   what's the reason behind disallowing get_user_pages() on VM_RESERVED 
> regions?  vmmon uses VM_RESERVED on its 'vma' as otherwise some kernels 
> used by SUSE complained loudly about mismatch between PageReserved() and 
> VM_RESERVED flags.
> 

Hi Petr,

The reason is that VM_RESERVED indicates that the core vm is not allowed
to touch any 'struct page' through this mapping, which get_user_pages
would do.

>   I'll remove it from vmmon for >= 2.6.14 kernels as that bogus test 
> never made to Linux kernel, but I cannot find any reason why 
> get_user_pages() should not work on VM_RESERVED (or VM_IO for that 
> matter) user pages.  Can you show me reasoning behind that decision ?

The reasoning behind the decision was so VM_RESERVED is usable for a
complete replacement to PageReserved. For example mappings through
/dev/mem should not touch the page count.

You may be able to go a step further and clear PageReserved from your
pages as well, and thus have a working driver without special casing
for both kernels.

Thanks,
Nick

-- 
SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com