Messages in this thread | | | Date | Wed, 23 Jan 2013 13:17:13 +0000 | From | Mel Gorman <> | Subject | Re: [PATCH 5/6] mm: Fold page->_last_nid into page->flags where possible |
| |
On Tue, Jan 22, 2013 at 02:46:59PM -0800, Andrew Morton wrote: > On Tue, 22 Jan 2013 17:12:41 +0000 > Mel Gorman <mgorman@suse.de> wrote: > > > From: Peter Zijlstra <a.p.zijlstra@chello.nl> > > > > page->_last_nid fits into page->flags on 64-bit. The unlikely 32-bit NUMA > > configuration with NUMA Balancing will still need an extra page field. > > As Peter notes "Completely dropping 32bit support for CONFIG_NUMA_BALANCING > > would simplify things, but it would also remove the warning if we grow > > enough 64bit only page-flags to push the last-cpu out." > > How much space remains in the 64-bit page->flags? >
Good question.
There are 19 free bits in my configuration but it's related to CONFIG_NODES_SHIFT which is 9 for me (512 nodes) and very heavily affected by options such as CONFIG_SPARSEMEM_VMEMMAP. Memory hot-remove does not work with CONFIG_SPARSEMEM_VMEMMAP and enterprise distribution configs may be taking the performance hit to enable memory hot-remove. If I disable this option to enable memory hot-remove then there are 0 free bits in page->flags.
Your milage will vary *considerably*.
In answering this question I remembered that mminit_loglevel is able to answer these sort of questions but only if it's updated properly. I'll post a follow-up patch.
> Was this the best possible use of the remaining space? >
Another good question and I do not have a good answer. There is a definite cost to having a larger struct page on large memory systems. The benefit to saving flags on 64-bit page->flags for potential future use is more intangiable.
> It's good that we can undo this later by flipping > LAST_NID_NOT_IN_PAGE_FLAGS. >
Yes and it generates a dirty warning if it's forced to use LAST_NID_NOT_IN_PAGE_FLAGS.
> > [mgorman@suse.de: Minor modifications] > > Signed-off-by: Mel Gorman <mgorman@suse.de> > > Several of these patches are missing signoffs (Peter and Hugh). >
In the case of Peter's patches, they changed enough that I couldn't preserve the signed-off-by. This happened for the NUMA balancing patches too. I preserved the "From" and I'm hoping he'll respond to add his Signed-off-by to these patches if he's ok with them.
In Hugh's case he did not add his signed-off-by because he was not sure whether there was a gremlin hidden in there. If there is, I was not able to find it. It's up to him whether he wants to put his signed-off-by on it but I preserved the "From:".
> > > > ... > > > > +static inline int page_last_nid(struct page *page) > > +{ > > + return (page->flags >> LAST_NID_PGSHIFT) & LAST_NID_MASK; > > +} > > + > > +static inline int page_xchg_last_nid(struct page *page, int nid) > > +{ > > + unsigned long old_flags, flags; > > + int last_nid; > > + > > + do { > > + old_flags = flags = page->flags; > > + last_nid = page_last_nid(page); > > + > > + flags &= ~(LAST_NID_MASK << LAST_NID_PGSHIFT); > > + flags |= (nid & LAST_NID_MASK) << LAST_NID_PGSHIFT; > > + } while (unlikely(cmpxchg(&page->flags, old_flags, flags) != old_flags)); > > + > > + return last_nid; > > +} > > + > > +static inline void reset_page_last_nid(struct page *page) > > +{ > > + page_xchg_last_nid(page, (1 << LAST_NID_SHIFT) - 1); > > +} > > page_xchg_last_nid() and reset_page_last_nid() are getting nuttily > large. Please investigate uninlining them? >
Will do.
> reset_page_last_nid() is poorly named. page_reset_last_nid() would be > better, and consistent. >
Will fix.
-- Mel Gorman SUSE Labs
| |