Re: Page Replacement in 2.6.10

[Rahul Iyer <rni@xxxxxxxxxxxxxx>]

Hi Rik,
In one of the earlier CART threads, you had mentioned that you had a way 
of representing each non resident page by a 32 bit value. Could you 
please elaborate on this? I haven't given too much thought on this, but 
i was thinking of a 64 bit value... upper 32 bits being pid and lower 32 
being the virtual address. Would this work?
Thanks
Rahul


Rik van Riel wrote:

> On Thu, 10 Mar 2005, Rahul Iyer wrote:
>
>  
>
> > Unfortunately, there is no pointer from struct page to the page table 
> > entry (struct pte_t) using the page, probably because there may be more 
> > than one. I then looked at rmap.h and rmap.c in 2.6.11. In those files i 
> > saw page_referenced(...). From what i gather, this function tests and 
> > clears the referenced flag for the given pages. Am I right?
> >    
>
> Yes.  The code in rmap.c provides all the infrastructure you
> need to check the referenced bit on a per physical page basis.
>
>  
>
> > Another question... what is ignore_token? As I saw from the source,
> >    
>
>  
>
> > I couldn't get the logic behind this bit.
> >    
>
> Please read the paper referenced from mm/thrash.c ;)
>
> Unfortunately the implementation in the upstream kernel was
> made by me in only a day and a half, and the code needs a
> lot of work to run smoothly under low loads.  The algorithm
> does seem to work very well under high loads, though ...
>
>  
>
> > P.S: Are all pages that are not backed by files (mmap'd) called anonymous
> > pages?
> >    
>
> Yes.
>
>  

--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/