Re: [RFC] Tagging of vmalloc pages for supporting the pmalloc allocator

[Jerome Glisse <jglisse@redhat.com>]

On Thu, Aug 03, 2017 at 03:55:50PM +0200, Michal Hocko wrote:
> On Thu 03-08-17 15:20:31, Igor Stoppa wrote:
> > On 03/08/17 14:48, Michal Hocko wrote:
> > > On Thu 03-08-17 13:11:45, Igor Stoppa wrote:
> > >> On 02/08/17 20:08, Jerome Glisse wrote:
> > >>> On Wed, Aug 02, 2017 at 06:14:28PM +0300, Igor Stoppa wrote:
> > 
> > [...]
> > 
> > >>>> from include/linux/mm_types.h:
> > >>>>
> > >>>> struct page {
> > >>>> ...
> > >>>>   union {
> > >>>>     unsigned long private;		/* Mapping-private opaque data:
> > >>>> 				 	 * usually used for buffer_heads
> > >>>> 					 * if PagePrivate set; used for
> > >>>> 					 * swp_entry_t if PageSwapCache;
> > >>>> 					 * indicates order in the buddy
> > >>>> 					 * system if PG_buddy is set.
> > >>>> 					 */
> > 
> > [...]
> > 
> > >> If the "Mapping-private" was dropped or somehow connected exclusively to
> > >> the cases listed in the comment, then I think it would be more clear
> > >> that the comment needs to be intended as related to mapping in certain
> > >> cases only.
> > >> But it is otherwise ok to use the "private" field for whatever purpose
> > >> it might be suitable, as long as it is not already in use.
> > > 
> > > I would recommend adding a new field into the enum...
> > 
> > s/enum/union/ ?
> > 
> > If not, I am not sure what is the enum that you are talking about.
> 
> yeah, fat fingers on my side
> 
> > 
> > [...]
> > 
> > >> But, to reply more specifically to your advice, yes, I think I could add
> > >> a flag to vm_struct and then retrieve its value, for the address being
> > >> processed, by passing through find_vm_area().
> > > 
> > > ... and you can store vm_struct pointer to the struct page there 
> > 
> > "there" as in the new field of the union?
> > btw, what would be a meaningful name, since "private" is already taken?
> > 
> > For simplicity, I'll use, for now, "private2"
> 
> why not explicit vm_area?
> 
> > > and you> won't need to do the slow find_vm_area. I haven't checked
> > very closely
> > > but this should be possible in principle. I guess other callers might
> > > benefit from this as well.
> > 
> > I am confused about this: if "private2" is a pointer, but when I get an
> > address, I do not even know if the address represents a valid pmalloc
> > page, how can i know when it's ok to dereference "private2"?
> 
> because you can make all pages which back vmalloc mappings have vm_area
> pointer set.

Note that i think this might break some device driver that use vmap()
i think some of them use private field to store device driver specific
informations. But there likely is an unuse field in struct page that
can be use for that.

Jérôme