VM_RESERVED vs vm_normal_page()

VM_RESERVED vs vm_normal_page()

[Benjamin Herrenschmidt]

Hi folks !

What is the logic regarding VM_RESERVED, and more specifically, why is
vm_normal_page() nor returning NULL for these ?

I have struct pages that are a bit special for things like SPE mappings
on Cell and I'd like to avoid a lot of the stuff the VM tries to do on
them, like rmap accounting, etc... In fact, for almost everything, the
semantics I want are vm_normal_page() to return NULL... I have struct
pages because for now I need do_no_page() to work, but I must absolutely
avoid things like getting swapped out etc...

Thus I looked at the logic in vm_normal_page() and it really sucks...
this is pretty much a "heuristic" to differenciate remap_page_range from
copy_on_write stuff ... gack.

What is VM_RESERVED for then ? Could I just use that ? I currently only
have VM_IO set on my SPE VMAs but I could add it. The current
implementation of vm_normal_page() doesn't test for it though, maybe it
should ?

I have still a problem though, in the case vm_normal_page() is made to
return NULL...

In that case, unmapping of pages will still cause them to be released
(tlb_* -> free_pages_and_swap_cache -> release_pages ->
put_page_testzero) but if you fork(), copy_one_pte() will not call an
additional get_page() when vm_normal_page() returns NULL... thus I fear
the page count will become bocus accross forks...

In the long run, I want to stop using struct page's for the SPU
registers and local store, once Nick's new stuff gets in, though I might
have a go at hacking something together before... but in the meantime,
I'd like to figure out what is the _safe_ way of having a VMA containing
PTEs mapping to struct pages that are _not_ normal memory and thus
aren't to be swapped out, freed, or anything like that.

Cheers,
Ben. 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: VM_RESERVED vs vm_normal_page()

[Robin Holt]

On Wed, Dec 13, 2006 at 03:37:57PM +1100, Benjamin Herrenschmidt wrote:
> Hi folks !
> 
> What is the logic regarding VM_RESERVED, and more specifically, why is
> vm_normal_page() nor returning NULL for these ?

Near as I could ever tell from the discussion on linux-mm, it is a page
which should not be dumped.  If you have a normal page in a mapping
which you don't want swapped out, the only way I could ever figure to
prevent that from happening is by doing an extra get_user_page() on it
to add a reference.

Thanks,
Robin

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>