ocfs2: allocate lockres hash pages in an array

This allows us to have a hash table greater than a single page which greatly
improves dlm performance on some tests.

Signed-off-by: Daniel Phillips <phillips@google.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

4 files changed, 46 insertions, 13 deletions

diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 87612819c13b..0378ddbc8a8c 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -37,7 +37,10 @@
 #define DLM_THREAD_SHUFFLE_INTERVAL    5     // flush everything every 5 passes
 #define DLM_THREAD_MS                  200   // flush at least every 200 ms
 
-#define DLM_HASH_BUCKETS     (PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_SIZE		(1 << 14)
+#define DLM_HASH_PAGES		(DLM_HASH_SIZE / PAGE_SIZE)
+#define DLM_BUCKETS_PER_PAGE	(PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_BUCKETS	(DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
 
 /* Intended to make it easier for us to switch out hash functions */
 #define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l)
@@ -88,7 +91,7 @@ enum dlm_ctxt_state {
 struct dlm_ctxt
 {
 	struct list_head list;
-	struct hlist_head *lockres_hash;
+	struct hlist_head **lockres_hash;
 	struct list_head dirty_list;
 	struct list_head purge_list;
 	struct list_head pending_asts;
@@ -135,6 +138,11 @@ struct dlm_ctxt
 	struct list_head	dlm_eviction_callbacks;
 };
 
+static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
+{
+	return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
+}
+
 /* these keventd work queue items are for less-frequently
  * called functions that cannot be directly called from the
  * net message handlers for some reason, usually because
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index c7eae5d3324e..e119e4ddf93f 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -136,7 +136,7 @@ void dlm_dump_lock_resources(struct dlm_ctxt *dlm)
 
 	spin_lock(&dlm->spinlock);
 	for (i=0; i<DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, iter, bucket, hash_node)
 			dlm_print_one_lock_resource(res);
 	}
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 595b1c7cbd1f..80b8cce9cf3c 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -49,6 +49,30 @@
 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
 #include "cluster/masklog.h"
 
+static void dlm_free_pagevec(void **vec, int pages)
+{
+	while (pages--)
+		free_page((unsigned long)vec[pages]);
+	kfree(vec);
+}
+
+static void **dlm_alloc_pagevec(int pages)
+{
+	void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
+	int i;
+
+	if (!vec)
+		return NULL;
+
+	for (i = 0; i < pages; i++)
+		if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
+			goto out_free;
+	return vec;
+out_free:
+	dlm_free_pagevec(vec, i);
+	return NULL;
+}
+
 /*
  *
  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
@@ -90,7 +114,7 @@ void __dlm_insert_lockres(struct dlm_ctxt *dlm,
 	assert_spin_locked(&dlm->spinlock);
 
 	q = &res->lockname;
-	bucket = &(dlm->lockres_hash[q->hash % DLM_HASH_BUCKETS]);
+	bucket = dlm_lockres_hash(dlm, q->hash);
 
 	/* get a reference for our hashtable */
 	dlm_lockres_get(res);
@@ -110,7 +134,8 @@ struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 
 	assert_spin_locked(&dlm->spinlock);
 
-	bucket = dlm->lockres_hash + full_name_hash(name, len) % DLM_HASH_BUCKETS;
+	bucket = dlm_lockres_hash(dlm, hash);
+
 	hlist_for_each(list, bucket) {
 		struct dlm_lock_resource *res = hlist_entry(list,
 			struct dlm_lock_resource, hash_node);
@@ -191,7 +216,7 @@ static int dlm_wait_on_domain_helper(const char *domain)
 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 {
 	if (dlm->lockres_hash)
-		free_page((unsigned long) dlm->lockres_hash);
+		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
 
 	if (dlm->name)
 		kfree(dlm->name);
@@ -301,8 +326,8 @@ static void dlm_migrate_all_locks(struct dlm_ctxt *dlm)
 restart:
 	spin_lock(&dlm->spinlock);
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		while (!hlist_empty(&dlm->lockres_hash[i])) {
-			res = hlist_entry(dlm->lockres_hash[i].first,
+		while (!hlist_empty(dlm_lockres_hash(dlm, i))) {
+			res = hlist_entry(dlm_lockres_hash(dlm, i)->first,
 					  struct dlm_lock_resource, hash_node);
 			/* need reference when manually grabbing lockres */
 			dlm_lockres_get(res);
@@ -1188,7 +1213,7 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
 		goto leave;
 	}
 
-	dlm->lockres_hash = (struct hlist_head *) __get_free_page(GFP_KERNEL);
+	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
 	if (!dlm->lockres_hash) {
 		mlog_errno(-ENOMEM);
 		kfree(dlm->name);
@@ -1197,8 +1222,8 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
 		goto leave;
 	}
 
-	for (i=0; i<DLM_HASH_BUCKETS; i++)
-		INIT_HLIST_HEAD(&dlm->lockres_hash[i]);
+	for (i = 0; i < DLM_HASH_BUCKETS; i++)
+		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
 
 	strcpy(dlm->name, domain);
 	dlm->key = key;
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
index 4f3d482a7299..bf2a02eca62e 100644
--- a/fs/ocfs2/dlm/dlmrecovery.c
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -1719,7 +1719,7 @@ static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
 	 * the RECOVERING state and set the owner
 	 * if necessary */
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, hash_iter, bucket, hash_node) {
 			if (res->state & DLM_LOCK_RES_RECOVERING) {
 				if (res->owner == dead_node) {
@@ -1884,7 +1884,7 @@ static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node)
 	 *    need to be fired as a result.
 	 */
 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
-		bucket = &(dlm->lockres_hash[i]);
+		bucket = dlm_lockres_hash(dlm, i);
 		hlist_for_each_entry(res, iter, bucket, hash_node) {
  			/* always prune any $RECOVERY entries for dead nodes,
  			 * otherwise hangs can occur during later recovery */