Re: [PATCH v3 5/5] mm: page_alloc: reduce unnecessary binary search in early_pfn_valid()

[Jia He]

On 3/28/2018 1:51 AM, Daniel Vacek Wrote:
> On Mon, Mar 26, 2018 at 5:02 AM, Jia He <hejianet@xxxxxxxxx> wrote:
> > Commit b92df1de5d28 ("mm: page_alloc: skip over regions of invalid pfns
> > where possible") optimized the loop in memmap_init_zone(). But there is
> > still some room for improvement. E.g. in early_pfn_valid(), if pfn and
> > pfn+1 are in the same memblock region, we can record the last returned
> > memblock region index and check check pfn++ is still in the same region.
> >
> > Currently it only improve the performance on arm64 and will have no
> > impact on other arches.
> >
> > Signed-off-by: Jia He <jia.he@xxxxxxxxxxxxxxxx>
> > ---
> >   arch/x86/include/asm/mmzone_32.h |  2 +-
> >   include/linux/mmzone.h           | 12 +++++++++---
> >   mm/page_alloc.c                  |  2 +-
> >   3 files changed, 11 insertions(+), 5 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/mmzone_32.h b/arch/x86/include/asm/mmzone_32.h
> > index 73d8dd1..329d3ba 100644
> > --- a/arch/x86/include/asm/mmzone_32.h
> > +++ b/arch/x86/include/asm/mmzone_32.h
> > @@ -49,7 +49,7 @@ static inline int pfn_valid(int pfn)
> >          return 0;
> >   }
> >
> > -#define early_pfn_valid(pfn)   pfn_valid((pfn))
> > +#define early_pfn_valid(pfn, last_region_idx)  pfn_valid((pfn))
> >
> >   #endif /* CONFIG_DISCONTIGMEM */
> >
> > diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> > index d797716..3a686af 100644
> > --- a/include/linux/mmzone.h
> > +++ b/include/linux/mmzone.h
> > @@ -1267,9 +1267,15 @@ static inline int pfn_present(unsigned long pfn)
> >   })
> >   #else
> >   #define pfn_to_nid(pfn)                (0)
> > -#endif
> > +#endif /*CONFIG_NUMA*/
> > +
> > +#ifdef CONFIG_HAVE_ARCH_PFN_VALID
> > +#define early_pfn_valid(pfn, last_region_idx) \
> > +                               pfn_valid_region(pfn, last_region_idx)
> > +#else
> > +#define early_pfn_valid(pfn, last_region_idx)  pfn_valid(pfn)
> > +#endif /*CONFIG_HAVE_ARCH_PFN_VALID*/
> >
> > -#define early_pfn_valid(pfn)   pfn_valid(pfn)
> >   void sparse_init(void);
> >   #else
> >   #define sparse_init()  do {} while (0)
> > @@ -1288,7 +1294,7 @@ struct mminit_pfnnid_cache {
> >   };
> >
> >   #ifndef early_pfn_valid
> > -#define early_pfn_valid(pfn)   (1)
> > +#define early_pfn_valid(pfn, last_region_idx)  (1)
> >   #endif
> >
> >   void memory_present(int nid, unsigned long start, unsigned long end);
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 0bb0274..debccf3 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -5484,7 +5484,7 @@ void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
> >                  if (context != MEMMAP_EARLY)
> >                          goto not_early;
> >
> > -               if (!early_pfn_valid(pfn)) {
> > +               if (!early_pfn_valid(pfn, &idx)) {
> >   #if (defined CONFIG_HAVE_MEMBLOCK) && (defined CONFIG_HAVE_ARCH_PFN_VALID)
> >                          /*
> >                           * Skip to the pfn preceding the next valid one (or
> > --
> > 2.7.4
> Hmm, what about making index global variable instead of changing all
> the prototypes? Similar to early_pfnnid_cache for example. Something
> like:
>
> #ifdef CONFIG_HAVE_ARCH_PFN_VALID
> extern int early_region_idx __meminitdata;
> #define early_pfn_valid(pfn) \
>                                 pfn_valid_region(pfn, &early_region_idx)
> #else
> #define early_pfn_valid(pfn)  pfn_valid(pfn)
> #endif /*CONFIG_HAVE_ARCH_PFN_VALID*/
>
> And move this to arch/arm64/include/asm/page.h ?
>
> --nX
Yes. ok with me

--
Cheers,
Jia