On 8/17/07, Mulyadi Santosa <
mulyadi.santosa@xxxxxxxxx> wrote:
Hello Akshay...
> while reading about page replacement algos, i have a doubt....
> whenever page table entries are set, pages will be accessed without
> knowledge of kernel. So how no. of references made to a page present in
> cache are found out?? (especially in frquency based algos)
>
>
This is how I understand the mechanism. If you access those pages
directly i.e using something like mov <address> $variable, of course
kernel won't recognize it and page count reference can not be adjusted.
IIRC, the best the MMU unit can do is simply turn the "accessed" bit on
in the related PTE.
Page count reference is adjusted when you access them via wrapper
functions. I am not able to tell the exact function names, but here is
example scenario. You mmap() a file... then some data chunks are brought
into page cache. Since this is the only process right now which accesses
them, page ref count is set as 0. Later, while still mmap()-ing them,
another process also want to mmap() the same file. This time, they are
found on page cache, thus page ref count are incremented by 1.
Minutes later, they are no longer needed by both processes. Unmap() is
called, releasing the pages from process address space's mapping. Then
page ref count are back to zero.
I hope I illustrate the situation correctly for you and enough to bring
enlightenment.
regards,
Mulyadi
