Dr. Werner Fink wrote:
...
> What's about age_page() in shrink_mmap() mm/filemap.c ... any suggestions
> or hints. Is there a fast way to get the age of the pte's?
Actually, I've been running a patch at home that completely removes
the use of PG_referenced - instead of checking if the bit is clear,
I test if page->age == 0. This way pages in the page cache get aged
according to the same sematics as normal pages, instead of the one bit
of 'age'. Note that there are constants in swapctl.h that hint to this
being intended. The question remains whether the extra CPU overhead is
worth it.
I've been thinking that it might be worthwhile separating the aging and
reaping process - perhaps finally creating a list of marked pages that
could be stolen as nescessary, which would solve most of the low-memory
problems that keep coming up (this pool could be kept at 1 meg+ on most
systems).
-ben