[PATCH] [BLOCK] cfq-iosched: change cfq io context linking from list to tree

On setups with many disks, we spend a considerable amount of time
looking up the process-disk mapping on each queue of io. Testing with
a NULL based block driver, this costs 40-50% reduction in throughput
for 1000 disks.

Signed-off-by: Jens Axboe <axboe@suse.de>

3 files changed, 114 insertions, 124 deletions

diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index bde40a6ae6..bb43a16776 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -1190,19 +1190,19 @@ cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio)
         return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT));
 }
 
-static void cfq_free_io_context(struct cfq_io_context *cic)
+static void cfq_free_io_context(struct io_context *ioc)
 {
         struct cfq_io_context *__cic;
-        struct list_head *entry, *next;
-        int freed = 1;
+        struct rb_node *n;
+        int freed = 0;
 
-        list_for_each_safe(entry, next, &cic->list) {
-                __cic = list_entry(entry, struct cfq_io_context, list);
+        while ((n = rb_first(&ioc->cic_root)) != NULL) {
+                __cic = rb_entry(n, struct cfq_io_context, rb_node);
+                rb_erase(&__cic->rb_node, &ioc->cic_root);
                 kmem_cache_free(cfq_ioc_pool, __cic);
                 freed++;
         }
 
-        kmem_cache_free(cfq_ioc_pool, cic);
         if (atomic_sub_and_test(freed, &ioc_count) && ioc_gone)
                 complete(ioc_gone);
 }
@@ -1210,8 +1210,7 @@ static void cfq_free_io_context(struct cfq_io_context *cic)
 static void cfq_trim(struct io_context *ioc)
 {
         ioc->set_ioprio = NULL;
-        if (ioc->cic)
-                cfq_free_io_context(ioc->cic);
+        cfq_free_io_context(ioc);
 }
 
 /*
@@ -1250,26 +1249,26 @@ static void cfq_exit_single_io_context(struct cfq_io_context *cic)
         spin_unlock(q->queue_lock);
 }
 
-static void cfq_exit_io_context(struct cfq_io_context *cic)
+static void cfq_exit_io_context(struct io_context *ioc)
 {
         struct cfq_io_context *__cic;
-        struct list_head *entry;
         unsigned long flags;
-
-        local_irq_save(flags);
+        struct rb_node *n;
 
         /*
          * put the reference this task is holding to the various queues
          */
-        read_lock(&cfq_exit_lock);
-        list_for_each(entry, &cic->list) {
-                __cic = list_entry(entry, struct cfq_io_context, list);
+        read_lock_irqsave(&cfq_exit_lock, flags);
+
+        n = rb_first(&ioc->cic_root);
+        while (n != NULL) {
+                __cic = rb_entry(n, struct cfq_io_context, rb_node);
+
                 cfq_exit_single_io_context(__cic);
+                n = rb_next(n);
         }
 
-        cfq_exit_single_io_context(cic);
-        read_unlock(&cfq_exit_lock);
-        local_irq_restore(flags);
+        read_unlock_irqrestore(&cfq_exit_lock, flags);
 }
 
 static struct cfq_io_context *
@@ -1278,10 +1277,10 @@ cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
         struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask);
 
         if (cic) {
-                INIT_LIST_HEAD(&cic->list);
+                RB_CLEAR(&cic->rb_node);
+                cic->key = NULL;
                 cic->cfqq[ASYNC] = NULL;
                 cic->cfqq[SYNC] = NULL;
-                cic->key = NULL;
                 cic->last_end_request = jiffies;
                 cic->ttime_total = 0;
                 cic->ttime_samples = 0;
@@ -1373,15 +1372,17 @@ static inline void changed_ioprio(struct cfq_io_context *cic)
 static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio)
 {
         struct cfq_io_context *cic;
+        struct rb_node *n;
 
         write_lock(&cfq_exit_lock);
 
-        cic = ioc->cic;
-
-        changed_ioprio(cic);
-
-        list_for_each_entry(cic, &cic->list, list)
+        n = rb_first(&ioc->cic_root);
+        while (n != NULL) {
+                cic = rb_entry(n, struct cfq_io_context, rb_node);
+ 
                 changed_ioprio(cic);
+                n = rb_next(n);
+        }
 
         write_unlock(&cfq_exit_lock);
 
@@ -1445,14 +1446,67 @@ out:
         return cfqq;
 }
 
+static struct cfq_io_context *
+cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
+{
+        struct rb_node *n = ioc->cic_root.rb_node;
+        struct cfq_io_context *cic;
+        void *key = cfqd;
+
+        while (n) {
+                cic = rb_entry(n, struct cfq_io_context, rb_node);
+
+                if (key < cic->key)
+                        n = n->rb_left;
+                else if (key > cic->key)
+                        n = n->rb_right;
+                else
+                        return cic;
+        }
+
+        return NULL;
+}
+
+static inline void
+cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
+             struct cfq_io_context *cic)
+{
+        struct rb_node **p = &ioc->cic_root.rb_node;
+        struct rb_node *parent = NULL;
+        struct cfq_io_context *__cic;
+
+        read_lock(&cfq_exit_lock);
+
+        cic->ioc = ioc;
+        cic->key = cfqd;
+
+        ioc->set_ioprio = cfq_ioc_set_ioprio;
+
+        while (*p) {
+                parent = *p;
+                __cic = rb_entry(parent, struct cfq_io_context, rb_node);
+
+                if (cic->key < __cic->key)
+                        p = &(*p)->rb_left;
+                else if (cic->key > __cic->key)
+                        p = &(*p)->rb_right;
+                else
+                        BUG();
+        }
+
+        rb_link_node(&cic->rb_node, parent, p);
+        rb_insert_color(&cic->rb_node, &ioc->cic_root);
+        list_add(&cic->queue_list, &cfqd->cic_list);
+        read_unlock(&cfq_exit_lock);
+}
+
 /*
  * Setup general io context and cfq io context. There can be several cfq
  * io contexts per general io context, if this process is doing io to more
- * than one device managed by cfq. Note that caller is holding a reference to
- * cfqq, so we don't need to worry about it disappearing
+ * than one device managed by cfq.
  */
 static struct cfq_io_context *
-cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
+cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
 {
         struct io_context *ioc = NULL;
         struct cfq_io_context *cic;
@@ -1463,88 +1517,15 @@ cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
         if (!ioc)
                 return NULL;
 
-restart:
-        if ((cic = ioc->cic) == NULL) {
-                cic = cfq_alloc_io_context(cfqd, gfp_mask);
-
-                if (cic == NULL)
-                        goto err;
-
-                /*
-                 * manually increment generic io_context usage count, it
+        cic = cfq_cic_rb_lookup(cfqd, ioc);
+        if (cic)
+                goto out;