lkml.org 
[lkml]   [2019]   [Oct]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] mm/page_alloc: Make alloc_gigantic_page() available for general use
On Wed 16-10-19 10:08:21, David Hildenbrand wrote:
> On 16.10.19 09:34, Anshuman Khandual wrote:
[...]
> > +static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn,
> > + unsigned long nr_pages)
> > +{
> > + unsigned long i, end_pfn = start_pfn + nr_pages;
> > + struct page *page;
> > +
> > + for (i = start_pfn; i < end_pfn; i++) {
> > + page = pfn_to_online_page(i);
> > + if (!page)
> > + return false;
> > +
> > + if (page_zone(page) != z)
> > + return false;
> > +
> > + if (PageReserved(page))
> > + return false;
> > +
> > + if (page_count(page) > 0)
> > + return false;
> > +
> > + if (PageHuge(page))
> > + return false;
> > + }
>
> We might still try to allocate a lot of ranges that contain unmovable data
> (we could avoid isolating a lot of page blocks in the first place). I'd love
> to see something like pfn_range_movable() (similar, but different to
> is_mem_section_removable(), which uses has_unmovable_pages()).

Just to make sure I understand. Do you want has_unmovable_pages to be
called inside pfn_range_valid_contig?
[...]
> > +struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
> > + int nid, nodemask_t *nodemask)
> > +{
> > + unsigned long ret, pfn, flags;
> > + struct zonelist *zonelist;
> > + struct zone *zone;
> > + struct zoneref *z;
> > +
> > + zonelist = node_zonelist(nid, gfp_mask);
> > + for_each_zone_zonelist_nodemask(zone, z, zonelist,
> > + gfp_zone(gfp_mask), nodemask) {
>
> One important part is to never use the MOVABLE zone here (otherwise
> unmovable data would end up on the movable zone). But I guess the caller is
> responsible for that (not pass GFP_MOVABLE) like gigantic pages do.

Well, if the caller uses GFP_MOVABLE then the movability should be
implemented in some form. If that is not the case then it is a bug on
the caller behalf.

> > + spin_lock_irqsave(&zone->lock, flags);
> > +
> > + pfn = ALIGN(zone->zone_start_pfn, nr_pages);
>
> This alignment does not make too much sense when allowing passing in !power
> of two orders. Maybe the caller should specify the requested alignment
> instead? Or should we enforce this to be aligned to make our life easier for
> now?

Are there any usecases that would require than the page alignment?
--
Michal Hocko
SUSE Labs

\
 
 \ /
  Last update: 2019-10-16 10:51    [W:0.123 / U:0.192 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site