kiobuf mapping with "mem=" boot option

Hi,

We have an application which uses a separate pool of memory not managed by linux (reserved using "mem=" option at boot command line) and a memory manager which manages this pool of memory in user space. The algorithm works like this:

a. use "mem=xM" option to boot command line b. mmap() the memory above "xM" till end of actual RAM in user space application.
c. the mmaped region is managed by a user space memory manager.

This works fine for the application.

1. Now we want to access a chunk of memory which we allocated using the user level memory manager in kernel space also. For this we tried kiobuf approach on the memory which was allocated in user space. But this approach doesn't seem to work on this mmaped memory (map_user_kiobuf() gives EFAULT: Bad Address). Is this valid behaviour?

2. Also, will we be wrong in making the assumption that the pages for memory above x Megs will never be swapped out?

If the answer to question 1 above is, that memory which is beyond x Megs (mem=xM) can not be used for kiobuf mapping, then is it ok if we directly access the user space address from kernel space (assuming the answer to question 2 is yes).

Any information in this regard will be of great help.

Thanks in advance.

__________________________________
Do you Yahoo!?
Yahoo! Mail - 50x more storage than other providers!
http://promotions.yahoo.com/new_mail