[PATCH] deadline-iosched: remove elevator private drq request type
authorJens Axboe <axboe@suse.de>
Thu, 13 Jul 2006 10:36:41 +0000 (12:36 +0200)
committerJens Axboe <axboe@nelson.home.kernel.dk>
Sat, 30 Sep 2006 18:27:00 +0000 (20:27 +0200)
A big win, we now save an allocation/free on each request! With the
previous rb/hash abstractions, we can just reuse queuelist/donelist
for the FIFO data and be done with it.

Signed-off-by: Jens Axboe <axboe@suse.de>
block/deadline-iosched.c

index 8300ba1f15a09841ed0beff8a1085a5af3d5c3fc..3b3b4414170ee1871c571b5ac023a312a5b00c88 100644 (file)
@@ -37,7 +37,7 @@ struct deadline_data {
        /*
         * next in sort order. read, write or both are NULL
         */
-       struct deadline_rq *next_drq[2];
+       struct request *next_rq[2];
        unsigned int batching;          /* number of sequential requests made */
        sector_t last_sector;           /* head position */
        unsigned int starved;           /* times reads have starved writes */
@@ -49,34 +49,14 @@ struct deadline_data {
        int fifo_batch;
        int writes_starved;
        int front_merges;
-
-       mempool_t *drq_pool;
 };
 
-/*
- * pre-request data.
- */
-struct deadline_rq {
-       struct request *request;
-
-       /*
-        * expire fifo
-        */
-       struct list_head fifo;
-       unsigned long expires;
-};
-
-static void deadline_move_request(struct deadline_data *dd, struct deadline_rq *drq);
-
-static kmem_cache_t *drq_pool;
-
-#define RQ_DATA(rq)    ((struct deadline_rq *) (rq)->elevator_private)
+static void deadline_move_request(struct deadline_data *, struct request *);
 
 #define RQ_RB_ROOT(dd, rq)     (&(dd)->sort_list[rq_data_dir((rq))])
-#define DRQ_RB_ROOT(dd, drq)   RQ_RB_ROOT((drq)->request)
 
 static void
-deadline_add_drq_rb(struct deadline_data *dd, struct request *rq)
+deadline_add_rq_rb(struct deadline_data *dd, struct request *rq)
 {
        struct rb_root *root = RQ_RB_ROOT(dd, rq);
        struct request *__alias;
@@ -84,45 +64,43 @@ deadline_add_drq_rb(struct deadline_data *dd, struct request *rq)
 retry:
        __alias = elv_rb_add(root, rq);
        if (unlikely(__alias)) {
-               deadline_move_request(dd, RQ_DATA(__alias));
+               deadline_move_request(dd, __alias);
                goto retry;
        }
 }
 
 static inline void
-deadline_del_drq_rb(struct deadline_data *dd, struct deadline_rq *drq)
+deadline_del_rq_rb(struct deadline_data *dd, struct request *rq)
 {
-       struct request *rq = drq->request;
        const int data_dir = rq_data_dir(rq);
 
-       if (dd->next_drq[data_dir] == drq) {
+       if (dd->next_rq[data_dir] == rq) {
                struct rb_node *rbnext = rb_next(&rq->rb_node);
 
-               dd->next_drq[data_dir] = NULL;
+               dd->next_rq[data_dir] = NULL;
                if (rbnext)
-                       dd->next_drq[data_dir] = RQ_DATA(rb_entry_rq(rbnext));
+                       dd->next_rq[data_dir] = rb_entry_rq(rbnext);
        }
 
        elv_rb_del(RQ_RB_ROOT(dd, rq), rq);
 }
 
 /*
- * add drq to rbtree and fifo
+ * add rq to rbtree and fifo
  */
 static void
 deadline_add_request(struct request_queue *q, struct request *rq)
 {
        struct deadline_data *dd = q->elevator->elevator_data;
-       struct deadline_rq *drq = RQ_DATA(rq);
-       const int data_dir = rq_data_dir(drq->request);
+       const int data_dir = rq_data_dir(rq);
 
-       deadline_add_drq_rb(dd, rq);
+       deadline_add_rq_rb(dd, rq);
 
        /*
         * set expire time (only used for reads) and add to fifo list
         */
-       drq->expires = jiffies + dd->fifo_expire[data_dir];
-       list_add_tail(&drq->fifo, &dd->fifo_list[data_dir]);
+       rq_set_fifo_time(rq, jiffies + dd->fifo_expire[data_dir]);
+       list_add_tail(&rq->queuelist, &dd->fifo_list[data_dir]);
 }
 
 /*
@@ -130,11 +108,10 @@ deadline_add_request(struct request_queue *q, struct request *rq)
  */
 static void deadline_remove_request(request_queue_t *q, struct request *rq)
 {
-       struct deadline_rq *drq = RQ_DATA(rq);
        struct deadline_data *dd = q->elevator->elevator_data;
 
-       list_del_init(&drq->fifo);
-       deadline_del_drq_rb(dd, drq);
+       rq_fifo_clear(rq);
+       deadline_del_rq_rb(dd, rq);
 }
 
 static int
@@ -177,7 +154,7 @@ static void deadline_merged_request(request_queue_t *q, struct request *req,
         */
        if (type == ELEVATOR_FRONT_MERGE) {
                elv_rb_del(RQ_RB_ROOT(dd, req), req);
-               deadline_add_drq_rb(dd, req);
+               deadline_add_rq_rb(dd, req);
        }
 }
 
@@ -185,20 +162,14 @@ static void
 deadline_merged_requests(request_queue_t *q, struct request *req,
                         struct request *next)
 {
-       struct deadline_rq *drq = RQ_DATA(req);
-       struct deadline_rq *dnext = RQ_DATA(next);
-
-       BUG_ON(!drq);
-       BUG_ON(!dnext);
-
        /*
-        * if dnext expires before drq, assign its expire time to drq
-        * and move into dnext position (dnext will be deleted) in fifo
+        * if next expires before rq, assign its expire time to rq
+        * and move into next position (next will be deleted) in fifo
         */
-       if (!list_empty(&drq->fifo) && !list_empty(&dnext->fifo)) {
-               if (time_before(dnext->expires, drq->expires)) {
-                       list_move(&drq->fifo, &dnext->fifo);
-                       drq->expires = dnext->expires;
+       if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) {
+               if (time_before(rq_fifo_time(next), rq_fifo_time(req))) {
+                       list_move(&req->queuelist, &next->queuelist);
+                       rq_set_fifo_time(req, rq_fifo_time(next));
                }
        }
 
@@ -212,53 +183,50 @@ deadline_merged_requests(request_queue_t *q, struct request *req,
  * move request from sort list to dispatch queue.
  */
 static inline void
-deadline_move_to_dispatch(struct deadline_data *dd, struct deadline_rq *drq)
+deadline_move_to_dispatch(struct deadline_data *dd, struct request *rq)
 {
-       request_queue_t *q = drq->request->q;
+       request_queue_t *q = rq->q;
 
-       deadline_remove_request(q, drq->request);
-       elv_dispatch_add_tail(q, drq->request);
+       deadline_remove_request(q, rq);
+       elv_dispatch_add_tail(q, rq);
 }
 
 /*
  * move an entry to dispatch queue
  */
 static void
-deadline_move_request(struct deadline_data *dd, struct deadline_rq *drq)
+deadline_move_request(struct deadline_data *dd, struct request *rq)
 {
-       struct request *rq = drq->request;
        const int data_dir = rq_data_dir(rq);
        struct rb_node *rbnext = rb_next(&rq->rb_node);
 
-       dd->next_drq[READ] = NULL;
-       dd->next_drq[WRITE] = NULL;
+       dd->next_rq[READ] = NULL;
+       dd->next_rq[WRITE] = NULL;
 
        if (rbnext)
-               dd->next_drq[data_dir] = RQ_DATA(rb_entry_rq(rbnext));
+               dd->next_rq[data_dir] = rb_entry_rq(rbnext);
        
-       dd->last_sector = drq->request->sector + drq->request->nr_sectors;
+       dd->last_sector = rq->sector + rq->nr_sectors;
 
        /*
         * take it off the sort and fifo list, move
         * to dispatch queue
         */
-       deadline_move_to_dispatch(dd, drq);
+       deadline_move_to_dispatch(dd, rq);
 }
 
-#define list_entry_fifo(ptr)   list_entry((ptr), struct deadline_rq, fifo)
-
 /*
  * deadline_check_fifo returns 0 if there are no expired reads on the fifo,
  * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir])
  */
 static inline int deadline_check_fifo(struct deadline_data *dd, int ddir)
 {
-       struct deadline_rq *drq = list_entry_fifo(dd->fifo_list[ddir].next);
+       struct request *rq = rq_entry_fifo(dd->fifo_list[ddir].next);
 
        /*
-        * drq is expired!
+        * rq is expired!
         */
-       if (time_after(jiffies, drq->expires))
+       if (time_after(jiffies, rq_fifo_time(rq)))
                return 1;
 
        return 0;
@@ -273,21 +241,21 @@ static int deadline_dispatch_requests(request_queue_t *q, int force)
        struct deadline_data *dd = q->elevator->elevator_data;
        const int reads = !list_empty(&dd->fifo_list[READ]);
        const int writes = !list_empty(&dd->fifo_list[WRITE]);
-       struct deadline_rq *drq;
+       struct request *rq;
        int data_dir;
 
        /*
         * batches are currently reads XOR writes
         */
-       if (dd->next_drq[WRITE])
-               drq = dd->next_drq[WRITE];
+       if (dd->next_rq[WRITE])
+               rq = dd->next_rq[WRITE];
        else
-               drq = dd->next_drq[READ];
+               rq = dd->next_rq[READ];
 
-       if (drq) {
+       if (rq) {
                /* we have a "next request" */
                
-               if (dd->last_sector != drq->request->sector)
+               if (dd->last_sector != rq->sector)
                        /* end the batch on a non sequential request */
                        dd->batching += dd->fifo_batch;
                
@@ -336,34 +304,33 @@ dispatch_find_request:
        if (deadline_check_fifo(dd, data_dir)) {
                /* An expired request exists - satisfy it */
                dd->batching = 0;
-               drq = list_entry_fifo(dd->fifo_list[data_dir].next);
+               rq = rq_entry_fifo(dd->fifo_list[data_dir].next);
                
-       } else if (dd->next_drq[data_dir]) {
+       } else if (dd->next_rq[data_dir]) {
                /*
                 * The last req was the same dir and we have a next request in
                 * sort order. No expired requests so continue on from here.
                 */
-               drq = dd->next_drq[data_dir];
+               rq = dd->next_rq[data_dir];
        } else {
-               struct rb_node *n;
-
+               struct rb_node *node;
                /*
                 * The last req was the other direction or we have run out of
                 * higher-sectored requests. Go back to the lowest sectored
                 * request (1 way elevator) and start a new batch.
                 */
                dd->batching = 0;
-               n = rb_first(&dd->sort_list[data_dir]);
-               if (n)
-                       drq = RQ_DATA(rb_entry_rq(n));
+               node = rb_first(&dd->sort_list[data_dir]);
+               if (node)
+                       rq = rb_entry_rq(node);
        }
 
 dispatch_request:
        /*
-        * drq is the selected appropriate request.
+        * rq is the selected appropriate request.
         */
        dd->batching++;
-       deadline_move_request(dd, drq);
+       deadline_move_request(dd, rq);
 
        return 1;
 }
@@ -383,33 +350,21 @@ static void deadline_exit_queue(elevator_t *e)
        BUG_ON(!list_empty(&dd->fifo_list[READ]));
        BUG_ON(!list_empty(&dd->fifo_list[WRITE]));
 
-       mempool_destroy(dd->drq_pool);
        kfree(dd);
 }
 
 /*
- * initialize elevator private data (deadline_data), and alloc a drq for
- * each request on the free lists
+ * initialize elevator private data (deadline_data).
  */
 static void *deadline_init_queue(request_queue_t *q, elevator_t *e)
 {
        struct deadline_data *dd;
 
-       if (!drq_pool)
-               return NULL;
-
        dd = kmalloc_node(sizeof(*dd), GFP_KERNEL, q->node);
        if (!dd)
                return NULL;
        memset(dd, 0, sizeof(*dd));
 
-       dd->drq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
-                                       mempool_free_slab, drq_pool, q->node);
-       if (!dd->drq_pool) {
-               kfree(dd);
-               return NULL;
-       }
-
        INIT_LIST_HEAD(&dd->fifo_list[READ]);
        INIT_LIST_HEAD(&dd->fifo_list[WRITE]);
        dd->sort_list[READ] = RB_ROOT;
@@ -422,36 +377,6 @@ static void *deadline_init_queue(request_queue_t *q, elevator_t *e)
        return dd;
 }
 
-static void deadline_put_request(request_queue_t *q, struct request *rq)
-{
-       struct deadline_data *dd = q->elevator->elevator_data;
-       struct deadline_rq *drq = RQ_DATA(rq);
-
-       mempool_free(drq, dd->drq_pool);
-       rq->elevator_private = NULL;
-}
-
-static int
-deadline_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
-                    gfp_t gfp_mask)
-{
-       struct deadline_data *dd = q->elevator->elevator_data;
-       struct deadline_rq *drq;
-
-       drq = mempool_alloc(dd->drq_pool, gfp_mask);
-       if (drq) {
-               memset(drq, 0, sizeof(*drq));
-               drq->request = rq;
-
-               INIT_LIST_HEAD(&drq->fifo);
-
-               rq->elevator_private = drq;
-               return 0;
-       }
-
-       return 1;
-}
-
 /*
  * sysfs parts below
  */
@@ -533,8 +458,6 @@ static struct elevator_type iosched_deadline = {
                .elevator_queue_empty_fn =      deadline_queue_empty,
                .elevator_former_req_fn =       elv_rb_former_request,
                .elevator_latter_req_fn =       elv_rb_latter_request,
-               .elevator_set_req_fn =          deadline_set_request,
-               .elevator_put_req_fn =          deadline_put_request,
                .elevator_init_fn =             deadline_init_queue,
                .elevator_exit_fn =             deadline_exit_queue,
        },
@@ -546,24 +469,11 @@ static struct elevator_type iosched_deadline = {
 
 static int __init deadline_init(void)
 {
-       int ret;
-
-       drq_pool = kmem_cache_create("deadline_drq", sizeof(struct deadline_rq),
-                                    0, 0, NULL, NULL);
-
-       if (!drq_pool)
-               return -ENOMEM;
-
-       ret = elv_register(&iosched_deadline);
-       if (ret)
-               kmem_cache_destroy(drq_pool);
-
-       return ret;
+       return elv_register(&iosched_deadline);
 }
 
 static void __exit deadline_exit(void)
 {
-       kmem_cache_destroy(drq_pool);
        elv_unregister(&iosched_deadline);
 }