Re: [RFC 1/3] zsmalloc: move pages_allocated to zs_pool

[Sergey Senozhatsky <sergey.senozhatsky@gmail.com>]

On (08/13/14 12:11), Dan Streetman wrote:
> >> > On Wed, Aug 13, 2014 at 10:14 AM, Sergey Senozhatsky
> >> > <sergey.senozhatsky@gmail.com> wrote:
> >> > > On (08/13/14 09:59), Dan Streetman wrote:
> >> > >> On Tue, Aug 5, 2014 at 4:02 AM, Minchan Kim <minchan@kernel.org> wrote:
> >> > >> > Pages_allocated has counted in size_class structure and when user
> >> > >> > want to see total_size_bytes, it gathers all of value from each
> >> > >> > size_class to report the sum.
> >> > >> >
> >> > >> > It's not bad if user don't see the value often but if user start
> >> > >> > to see the value frequently, it would be not a good deal for
> >> > >> > performance POV.
> >> > >> >
> >> > >> > This patch moves the variable from size_class to zs_pool so it would
> >> > >> > reduce memory footprint (from [255 * 8byte] to [sizeof(atomic_t)])
> >> > >> > but it adds new locking overhead but it wouldn't be severe because
> >> > >> > it's not a hot path in zs_malloc(ie, it is called only when new
> >> > >> > zspage is created, not a object).
> >> > >>
> >> > >> Would using an atomic64_t without locking be simpler?
> >> > >
> >> > > it would be racy.
> >> >
> >> > oh.  atomic operations aren't smp safe?  is that because other
> >> > processors might use a stale value, and barriers must be added?  I
> >> > guess I don't quite understand the value of atomic then. :-/
> >>
> >> pool not only set the value, it also read it and make some decisions
> >> based on that value:
> >>
> >>       pages_allocated += X
> >>       if (pages_allocated >= max_pages_allocated)
> >>               return 0;
> >
> 
> I'm missing where that is?  I don't see that in this patch?
> 
> >
> > I mean, suppose this happens on two CPUs
> >
> > max_pages_allocated is 10; current pages_allocated is 8. now you have 2 zs_malloc()
> > happenning on two CPUs. each of them will do `pages_allocated += 1'. the problem is
> > that both will see 10 at `if (pages_allocated >= max_pages_allocated)', so we will
> > fail 2 operations, while we only were supposed to fail one.
> 
> Do you mean this from the 2/3 patch:

yeah. sorry for being unclear, I was really sleepy.

> @@ -946,6 +947,8 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size)
>                 set_zspage_mapping(first_page, class->index, ZS_EMPTY);
>                 spin_lock(&pool->stat_lock);
>                 pool->pages_allocated += class->pages_per_zspage;
> +               if (pool->max_pages_allocated < pool->pages_allocated)
> +                       pool->max_pages_allocated = pool->pages_allocated;
>                 spin_unlock(&pool->stat_lock);
>                 spin_lock(&class->lock);
>         }
> 
> I see, yeah the max > allocated check before setting is easiest done
> with a spinlock.  I think pages_allocated could still be done as
> atomic, just using atomic_add_return() to grab the current value to
> check against, but keeping them the same type and both protected by
> the same spinlock I guess simplifies things.  Although, if they were
> both atomic, then the *only* place that would need a spinlock would be
> this check - reading the (atomic) max_pages_allocated wouldn't need a
> spinlock, nor would clearing it to 0.

makes sense.

	-ss

> >> > >>
> >> > >> >
> >> > >> > Signed-off-by: Minchan Kim <minchan@kernel.org>
> >> > >> > ---
> >> > >> >  mm/zsmalloc.c | 30 ++++++++++++++++--------------
> >> > >> >  1 file changed, 16 insertions(+), 14 deletions(-)
> >> > >> >
> >> > >> > diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
> >> > >> > index fe78189624cf..a6089bd26621 100644
> >> > >> > --- a/mm/zsmalloc.c
> >> > >> > +++ b/mm/zsmalloc.c
> >> > >> > @@ -198,9 +198,6 @@ struct size_class {
> >> > >> >
> >> > >> >         spinlock_t lock;
> >> > >> >
> >> > >> > -       /* stats */
> >> > >> > -       u64 pages_allocated;
> >> > >> > -
> >> > >> >         struct page *fullness_list[_ZS_NR_FULLNESS_GROUPS];
> >> > >> >  };
> >> > >> >
> >> > >> > @@ -216,9 +213,12 @@ struct link_free {
> >> > >> >  };
> >> > >> >
> >> > >> >  struct zs_pool {
> >> > >> > +       spinlock_t stat_lock;
> >> > >> > +
> >> > >> >         struct size_class size_class[ZS_SIZE_CLASSES];
> >> > >> >
> >> > >> >         gfp_t flags;    /* allocation flags used when growing pool */
> >> > >> > +       unsigned long pages_allocated;
> >> > >> >  };
> >> > >> >
> >> > >> >  /*
> >> > >> > @@ -882,6 +882,7 @@ struct zs_pool *zs_create_pool(gfp_t flags)
> >> > >> >
> >> > >> >         }
> >> > >> >
> >> > >> > +       spin_lock_init(&pool->stat_lock);
> >> > >> >         pool->flags = flags;
> >> > >> >
> >> > >> >         return pool;
> >> > >> > @@ -943,8 +944,10 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size)
> >> > >> >                         return 0;
> >> > >> >
> >> > >> >                 set_zspage_mapping(first_page, class->index, ZS_EMPTY);
> >> > >> > +               spin_lock(&pool->stat_lock);
> >> > >> > +               pool->pages_allocated += class->pages_per_zspage;
> >> > >> > +               spin_unlock(&pool->stat_lock);
> >> > >> >                 spin_lock(&class->lock);
> >> > >> > -               class->pages_allocated += class->pages_per_zspage;
> >> > >> >         }
> >> > >> >
> >> > >> >         obj = (unsigned long)first_page->freelist;
> >> > >> > @@ -997,14 +1000,14 @@ void zs_free(struct zs_pool *pool, unsigned long obj)
> >> > >> >
> >> > >> >         first_page->inuse--;
> >> > >> >         fullness = fix_fullness_group(pool, first_page);
> >> > >> > -
> >> > >> > -       if (fullness == ZS_EMPTY)
> >> > >> > -               class->pages_allocated -= class->pages_per_zspage;
> >> > >> > -
> >> > >> >         spin_unlock(&class->lock);
> >> > >> >
> >> > >> > -       if (fullness == ZS_EMPTY)
> >> > >> > +       if (fullness == ZS_EMPTY) {
> >> > >> > +               spin_lock(&pool->stat_lock);
> >> > >> > +               pool->pages_allocated -= class->pages_per_zspage;
> >> > >> > +               spin_unlock(&pool->stat_lock);
> >> > >> >                 free_zspage(first_page);
> >> > >> > +       }
> >> > >> >  }
> >> > >> >  EXPORT_SYMBOL_GPL(zs_free);
> >> > >> >
> >> > >> > @@ -1100,12 +1103,11 @@ EXPORT_SYMBOL_GPL(zs_unmap_object);
> >> > >> >
> >> > >> >  u64 zs_get_total_size_bytes(struct zs_pool *pool)
> >> > >> >  {
> >> > >> > -       int i;
> >> > >> > -       u64 npages = 0;
> >> > >> > -
> >> > >> > -       for (i = 0; i < ZS_SIZE_CLASSES; i++)
> >> > >> > -               npages += pool->size_class[i].pages_allocated;
> >> > >> > +       u64 npages;
> >> > >> >
> >> > >> > +       spin_lock(&pool->stat_lock);
> >> > >> > +       npages = pool->pages_allocated;
> >> > >> > +       spin_unlock(&pool->stat_lock);
> >> > >> >         return npages << PAGE_SHIFT;
> >> > >> >  }
> >> > >> >  EXPORT_SYMBOL_GPL(zs_get_total_size_bytes);
> >> > >> > --
> >> > >> > 2.0.0
> >> > >> >
> >> > >> > --
> >> > >> > 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>
> >> > >>
> >> >
> >>
> 

--
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>