On Fri, 21 Sep 2007 18:02:47 +0100 (BST)
Hugh Dickins <hugh@veritas.com> wrote:
quoted text
> > 3. I want to *try* page->mapping overriding... store  memory resource controller's   
> >    information in page->mapping. By this, memory controller doesn't enlarge sizeof
> >    struct page. (works well in my small test.)
> >    Before doing that, I have to hide page->mapping from direct access.
> 
> My own vote (nothing more) would be for you to set this aside until
> some future time when there aren't a dozen developers all trampling
> over each other in this area.
> 
> They're invasive little changes affecting all filesystems, whereas what
> we've done so far with page->mapping hasn't affected filesystems at all.
> 
I found that each FS doesn't touch page->mapping so much as I expected.
(except for ReiserFS)
But ok, I admit changing this will confuse people.

quoted text
> 3: well, saving memory is good, but I think it could wait until some
> other time, particularly since the memory controller isn't in yet.
> 
Yes, if extra field in page struct is not hazard to push memory controller,
I don't have much motivation. 
Because extra 8 bytes makes page struct to be 64 bytes(in 64bit), extra 8 bytes
is the last space, I think.

quoted text
> If we were to attack page->mapping to save memory from struct page,
> then we should consider Magnus Damm's idea too: he suggested it could
> be replaced by a pointer to the radixtree slot (something else needed
> in the anon case), from which "index" could be deduced via alignment
> instead of keeping it in struct page (details to be filled in ...)
> 
There is a bit difference. My purpose is "avoid making struct page larger",
not "making struct page smaller". 

quoted text
> Or should I now leave PG_swapcache as is,
> given your designs on page->mapping?
> 
 will conflict with my idea ?
==
http://marc.info/?l=linux-mm&m=118956492926821&w=2
==

Anyway, I'm not in hurry about this patch-set. I'll see what memory controller
will go. Other people seems to have an idea to implement 
pfn <-> container_info_per_page function.
(But this kind of function is not welcomed always.)

Thank you for comments.

quoted text
> p.s. Sorry to niggle, but next time, please say [PATCH 1/3] etc.
> rather than [PATCH] Long Description [1/3], so it's easier to
> sort the mail subjects by eye in limited columns - thanks.
> 
sorry, I'll consider well next time.

Thanks,
-Kame
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/