Thread (26 messages) 26 messages, 6 authors, 2014-08-17

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

From: Sergey Senozhatsky <hidden>
Date: 2014-08-14 13:04:56
Also in: lkml

On (08/13/14 12:11), Dan Streetman wrote:
quoted
quoted
quoted
On Wed, Aug 13, 2014 at 10:14 AM, Sergey Senozhatsky
[off-list ref] wrote:
quoted
On (08/13/14 09:59), Dan Streetman wrote:
quoted
On Tue, Aug 5, 2014 at 4:02 AM, Minchan Kim [off-list ref] wrote:
quoted
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?
quoted
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.
quoted hunk ↗ jump to hunk
@@ -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
quoted
quoted
quoted
quoted
quoted
quoted
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>
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help