staging: zcache: remove zcache_direct_reclaim_lock
authorSeth Jennings <sjenning@linux.vnet.ibm.com>
Wed, 12 Oct 2011 19:41:00 +0000 (14:41 -0500)
committerGreg Kroah-Hartman <gregkh@suse.de>
Mon, 17 Oct 2011 22:24:11 +0000 (15:24 -0700)
zcache_do_preload() currently does a spin_trylock() on the
zcache_direct_reclaim_lock. Holding this lock intends to prevent
shrink_zcache_memory() from evicting zbud pages as a result
of a preload.

However, it also prevents two threads from
executing zcache_do_preload() at the same time.  The first
thread will obtain the lock and the second thread's spin_trylock()
will fail (an aborted preload) causing the page to be either lost
(cleancache) or pushed out to the swap device (frontswap). It
also doesn't ensure that the call to shrink_zcache_memory() is
on the same thread as the call to zcache_do_preload().

Additional, there is no need for this mechanism because all
zcache_do_preload() calls that come down from cleancache already
have PF_MEMALLOC set in the process flags which prevents
direct reclaim in the memory manager. If the zcache_do_preload()
call is done from the frontswap path, we _want_ reclaim to be
done (which it isn't right now).

This patch removes the zcache_direct_reclaim_lock and related
statistics in zcache.

Based on v3.1-rc8

Signed-off-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Reviewed-by: Dave Hansen <dave@linux.vnet.ibm.com>
Acked-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/staging/zcache/zcache-main.c

index 22b0a75..3768bf2 100644 (file)
@@ -962,15 +962,6 @@ out:
 static unsigned long zcache_failed_get_free_pages;
 static unsigned long zcache_failed_alloc;
 static unsigned long zcache_put_to_flush;
-static unsigned long zcache_aborted_preload;
-static unsigned long zcache_aborted_shrink;
-
-/*
- * Ensure that memory allocation requests in zcache don't result
- * in direct reclaim requests via the shrinker, which would cause
- * an infinite loop.  Maybe a GFP flag would be better?
- */
-static DEFINE_SPINLOCK(zcache_direct_reclaim_lock);
 
 /*
  * for now, used named slabs so can easily track usage; later can
@@ -1009,10 +1000,6 @@ static int zcache_do_preload(struct tmem_pool *pool)
                goto out;
        if (unlikely(zcache_obj_cache == NULL))
                goto out;
-       if (!spin_trylock(&zcache_direct_reclaim_lock)) {
-               zcache_aborted_preload++;
-               goto out;
-       }
        preempt_disable();
        kp = &__get_cpu_var(zcache_preloads);
        while (kp->nr < ARRAY_SIZE(kp->objnodes)) {
@@ -1021,7 +1008,7 @@ static int zcache_do_preload(struct tmem_pool *pool)
                                ZCACHE_GFP_MASK);
                if (unlikely(objnode == NULL)) {
                        zcache_failed_alloc++;
-                       goto unlock_out;
+                       goto out;
                }
                preempt_disable();
                kp = &__get_cpu_var(zcache_preloads);
@@ -1034,13 +1021,13 @@ static int zcache_do_preload(struct tmem_pool *pool)
        obj = kmem_cache_alloc(zcache_obj_cache, ZCACHE_GFP_MASK);
        if (unlikely(obj == NULL)) {
                zcache_failed_alloc++;
-               goto unlock_out;
+               goto out;
        }
        page = (void *)__get_free_page(ZCACHE_GFP_MASK);
        if (unlikely(page == NULL)) {
                zcache_failed_get_free_pages++;
                kmem_cache_free(zcache_obj_cache, obj);
-               goto unlock_out;
+               goto out;
        }
        preempt_disable();
        kp = &__get_cpu_var(zcache_preloads);
@@ -1053,8 +1040,6 @@ static int zcache_do_preload(struct tmem_pool *pool)
        else
                free_page((unsigned long)page);
        ret = 0;
-unlock_out:
-       spin_unlock(&zcache_direct_reclaim_lock);
 out:
        return ret;
 }
@@ -1429,8 +1414,6 @@ ZCACHE_SYSFS_RO(evicted_buddied_pages);
 ZCACHE_SYSFS_RO(failed_get_free_pages);
 ZCACHE_SYSFS_RO(failed_alloc);
 ZCACHE_SYSFS_RO(put_to_flush);
-ZCACHE_SYSFS_RO(aborted_preload);
-ZCACHE_SYSFS_RO(aborted_shrink);
 ZCACHE_SYSFS_RO(compress_poor);
 ZCACHE_SYSFS_RO(mean_compress_poor);
 ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_raw_pages);
@@ -1472,8 +1455,6 @@ static struct attribute *zcache_attrs[] = {
        &zcache_failed_get_free_pages_attr.attr,
        &zcache_failed_alloc_attr.attr,
        &zcache_put_to_flush_attr.attr,
-       &zcache_aborted_preload_attr.attr,
-       &zcache_aborted_shrink_attr.attr,
        &zcache_zbud_unbuddied_list_counts_attr.attr,
        &zcache_zbud_cumul_chunk_counts_attr.attr,
        &zcache_zv_curr_dist_counts_attr.attr,
@@ -1513,11 +1494,7 @@ static int shrink_zcache_memory(struct shrinker *shrink,
                if (!(gfp_mask & __GFP_FS))
                        /* does this case really need to be skipped? */
                        goto out;
-               if (spin_trylock(&zcache_direct_reclaim_lock)) {
-                       zbud_evict_pages(nr);
-                       spin_unlock(&zcache_direct_reclaim_lock);
-               } else
-                       zcache_aborted_shrink++;
+               zbud_evict_pages(nr);
        }
        ret = (int)atomic_read(&zcache_zbud_curr_raw_pages);
 out: