aboutsummaryrefslogtreecommitdiff
path: root/drivers/block
diff options
context:
space:
mode:
authorLinus Torvalds2005-10-28 08:56:34 -0700
committerLinus Torvalds2005-10-28 08:56:34 -0700
commit5dd962494f76fb3ef1196cd420b5d6260d7a3766 (patch)
treef6eda4fcea880e420420a2886b84ad35a6b00f28 /drivers/block
parent28d721e24c88496ff8e9c4a0959bdc1415c0658e (diff)
parent772eca7825358f52cf208afd7ab8ce22722ce41a (diff)
Merge branch 'elevator-switch' of git://brick.kernel.dk/data/git/linux-2.6-block
Manual fixup for trivial "gfp_t" changes.
Diffstat (limited to 'drivers/block')
-rw-r--r--drivers/block/elevator.c77
-rw-r--r--drivers/block/ll_rw_blk.c142
2 files changed, 61 insertions, 158 deletions
diff --git a/drivers/block/elevator.c b/drivers/block/elevator.c
index 3b8099ccedff..55621d5c5774 100644
--- a/drivers/block/elevator.c
+++ b/drivers/block/elevator.c
@@ -34,6 +34,7 @@
#include <linux/slab.h>
#include <linux/init.h>
#include <linux/compiler.h>
+#include <linux/delay.h>
#include <asm/uaccess.h>
@@ -131,11 +132,7 @@ static int elevator_attach(request_queue_t *q, struct elevator_type *e,
eq->ops = &e->ops;
eq->elevator_type = e;
- INIT_LIST_HEAD(&q->queue_head);
- q->last_merge = NULL;
q->elevator = eq;
- q->end_sector = 0;
- q->boundary_rq = NULL;
if (eq->ops->elevator_init_fn)
ret = eq->ops->elevator_init_fn(q, eq);
@@ -184,6 +181,11 @@ int elevator_init(request_queue_t *q, char *name)
struct elevator_queue *eq;
int ret = 0;
+ INIT_LIST_HEAD(&q->queue_head);
+ q->last_merge = NULL;
+ q->end_sector = 0;
+ q->boundary_rq = NULL;
+
elevator_setup_default();
if (!name)
@@ -336,23 +338,14 @@ void __elv_add_request(request_queue_t *q, struct request *rq, int where,
q->end_sector = rq_end_sector(rq);
q->boundary_rq = rq;
}
- }
+ } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
+ where = ELEVATOR_INSERT_BACK;
if (plug)
blk_plug_device(q);
rq->q = q;
- if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) {
- /*
- * if drain is set, store the request "locally". when the drain
- * is finished, the requests will be handed ordered to the io
- * scheduler
- */
- list_add_tail(&rq->queuelist, &q->drain_list);
- return;
- }
-
switch (where) {
case ELEVATOR_INSERT_FRONT:
rq->flags |= REQ_SOFTBARRIER;
@@ -659,25 +652,36 @@ EXPORT_SYMBOL_GPL(elv_unregister);
* switch to new_e io scheduler. be careful not to introduce deadlocks -
* we don't free the old io scheduler, before we have allocated what we
* need for the new one. this way we have a chance of going back to the old
- * one, if the new one fails init for some reason. we also do an intermediate
- * switch to noop to ensure safety with stack-allocated requests, since they
- * don't originate from the block layer allocator. noop is safe here, because
- * it never needs to touch the elevator itself for completion events. DRAIN
- * flags will make sure we don't touch it for additions either.
+ * one, if the new one fails init for some reason.
*/
static void elevator_switch(request_queue_t *q, struct elevator_type *new_e)
{
- elevator_t *e = kmalloc(sizeof(elevator_t), GFP_KERNEL);
- struct elevator_type *noop_elevator = NULL;
- elevator_t *old_elevator;
+ elevator_t *old_elevator, *e;
+ /*
+ * Allocate new elevator
+ */
+ e = kmalloc(sizeof(elevator_t), GFP_KERNEL);
if (!e)
goto error;
/*
- * first step, drain requests from the block freelist
+ * Turn on BYPASS and drain all requests w/ elevator private data
*/
- blk_wait_queue_drained(q, 0);
+ spin_lock_irq(q->queue_lock);
+
+ set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
+
+ while (q->elevator->ops->elevator_dispatch_fn(q, 1))
+ ;
+
+ while (q->rq.elvpriv) {
+ spin_unlock_irq(q->queue_lock);
+ msleep(10);
+ spin_lock_irq(q->queue_lock);
+ }
+
+ spin_unlock_irq(q->queue_lock);
/*
* unregister old elevator data
@@ -686,18 +690,6 @@ static void elevator_switch(request_queue_t *q, struct elevator_type *new_e)
old_elevator = q->elevator;
/*
- * next step, switch to noop since it uses no private rq structures
- * and doesn't allocate any memory for anything. then wait for any
- * non-fs requests in-flight
- */
- noop_elevator = elevator_get("noop");
- spin_lock_irq(q->queue_lock);
- elevator_attach(q, noop_elevator, e);
- spin_unlock_irq(q->queue_lock);
-
- blk_wait_queue_drained(q, 1);
-
- /*
* attach and start new elevator
*/
if (elevator_attach(q, new_e, e))
@@ -707,11 +699,10 @@ static void elevator_switch(request_queue_t *q, struct elevator_type *new_e)
goto fail_register;
/*
- * finally exit old elevator and start queue again
+ * finally exit old elevator and turn off BYPASS.
*/
elevator_exit(old_elevator);
- blk_finish_queue_drain(q);
- elevator_put(noop_elevator);
+ clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
return;
fail_register:
@@ -720,13 +711,13 @@ fail_register:
* one again (along with re-adding the sysfs dir)
*/
elevator_exit(e);
+ e = NULL;
fail:
q->elevator = old_elevator;
elv_register_queue(q);
- blk_finish_queue_drain(q);
+ clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
+ kfree(e);
error:
- if (noop_elevator)
- elevator_put(noop_elevator);
elevator_put(new_e);
printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name);
}
diff --git a/drivers/block/ll_rw_blk.c b/drivers/block/ll_rw_blk.c
index c4f186223bc5..0af73512b9a8 100644
--- a/drivers/block/ll_rw_blk.c
+++ b/drivers/block/ll_rw_blk.c
@@ -263,8 +263,6 @@ void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
blk_queue_activity_fn(q, NULL, NULL);
-
- INIT_LIST_HEAD(&q->drain_list);
}
EXPORT_SYMBOL(blk_queue_make_request);
@@ -1050,6 +1048,7 @@ static char *rq_flags[] = {
"REQ_STARTED",
"REQ_DONTPREP",
"REQ_QUEUED",
+ "REQ_ELVPRIV",
"REQ_PC",
"REQ_BLOCK_PC",
"REQ_SENSE",
@@ -1640,9 +1639,9 @@ static int blk_init_free_list(request_queue_t *q)
rl->count[READ] = rl->count[WRITE] = 0;
rl->starved[READ] = rl->starved[WRITE] = 0;
+ rl->elvpriv = 0;
init_waitqueue_head(&rl->wait[READ]);
init_waitqueue_head(&rl->wait[WRITE]);
- init_waitqueue_head(&rl->drain);
rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
mempool_free_slab, request_cachep, q->node);
@@ -1785,12 +1784,14 @@ EXPORT_SYMBOL(blk_get_queue);
static inline void blk_free_request(request_queue_t *q, struct request *rq)
{
- elv_put_request(q, rq);
+ if (rq->flags & REQ_ELVPRIV)
+ elv_put_request(q, rq);
mempool_free(rq, q->rq.rq_pool);
}
static inline struct request *
-blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, gfp_t gfp_mask)
+blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
+ int priv, gfp_t gfp_mask)
{
struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
@@ -1803,11 +1804,15 @@ blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, gfp_t gfp_mask)
*/
rq->flags = rw;
- if (!elv_set_request(q, rq, bio, gfp_mask))
- return rq;
+ if (priv) {
+ if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
+ mempool_free(rq, q->rq.rq_pool);
+ return NULL;
+ }
+ rq->flags |= REQ_ELVPRIV;
+ }
- mempool_free(rq, q->rq.rq_pool);
- return NULL;
+ return rq;
}
/*
@@ -1863,22 +1868,18 @@ static void __freed_request(request_queue_t *q, int rw)
* A request has just been released. Account for it, update the full and
* congestion status, wake up any waiters. Called under q->queue_lock.
*/
-static void freed_request(request_queue_t *q, int rw)
+static void freed_request(request_queue_t *q, int rw, int priv)
{
struct request_list *rl = &q->rq;
rl->count[rw]--;
+ if (priv)
+ rl->elvpriv--;
__freed_request(q, rw);
if (unlikely(rl->starved[rw ^ 1]))
__freed_request(q, rw ^ 1);
-
- if (!rl->count[READ] && !rl->count[WRITE]) {
- smp_mb();
- if (unlikely(waitqueue_active(&rl->drain)))
- wake_up(&rl->drain);
- }
}
#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
@@ -1893,9 +1894,7 @@ static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
struct request *rq = NULL;
struct request_list *rl = &q->rq;
struct io_context *ioc = current_io_context(GFP_ATOMIC);
-
- if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)))
- goto out;
+ int priv;
if (rl->count[rw]+1 >= q->nr_requests) {
/*
@@ -1940,9 +1939,14 @@ get_rq:
rl->starved[rw] = 0;
if (rl->count[rw] >= queue_congestion_on_threshold(q))
set_queue_congested(q, rw);
+
+ priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
+ if (priv)
+ rl->elvpriv++;
+
spin_unlock_irq(q->queue_lock);
- rq = blk_alloc_request(q, rw, bio, gfp_mask);
+ rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
if (!rq) {
/*
* Allocation failed presumably due to memory. Undo anything
@@ -1952,7 +1956,7 @@ get_rq:
* wait queue, but this is pretty rare.
*/
spin_lock_irq(q->queue_lock);
- freed_request(q, rw);
+ freed_request(q, rw, priv);
/*
* in the very unlikely event that allocation failed and no
@@ -2470,11 +2474,12 @@ static void __blk_put_request(request_queue_t *q, struct request *req)
*/
if (rl) {
int rw = rq_data_dir(req);
+ int priv = req->flags & REQ_ELVPRIV;
BUG_ON(!list_empty(&req->queuelist));
blk_free_request(q, req);
- freed_request(q, rw);
+ freed_request(q, rw, priv);
}
}
@@ -2802,97 +2807,6 @@ static inline void blk_partition_remap(struct bio *bio)
}
}
-void blk_finish_queue_drain(request_queue_t *q)
-{
- struct request_list *rl = &q->rq;
- struct request *rq;
- int requeued = 0;
-
- spin_lock_irq(q->queue_lock);
- clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
-
- while (!list_empty(&q->drain_list)) {
- rq = list_entry_rq(q->drain_list.next);
-
- list_del_init(&rq->queuelist);
- elv_requeue_request(q, rq);
- requeued++;
- }
-
- if (requeued)
- q->request_fn(q);
-
- spin_unlock_irq(q->queue_lock);
-
- wake_up(&rl->wait[0]);
- wake_up(&rl->wait[1]);
- wake_up(&rl->drain);
-}
-
-static int wait_drain(request_queue_t *q, struct request_list *rl, int dispatch)
-{
- int wait = rl->count[READ] + rl->count[WRITE];
-
- if (dispatch)
- wait += !list_empty(&q->queue_head);
-
- return wait;
-}
-
-/*
- * We rely on the fact that only requests allocated through blk_alloc_request()
- * have io scheduler private data structures associated with them. Any other
- * type of request (allocated on stack or through kmalloc()) should not go
- * to the io scheduler core, but be attached to the queue head instead.
- */
-void blk_wait_queue_drained(request_queue_t *q, int wait_dispatch)
-{
- struct request_list *rl = &q->rq;
- DEFINE_WAIT(wait);
-
- spin_lock_irq(q->queue_lock);
- set_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
-
- while (wait_drain(q, rl, wait_dispatch)) {
- prepare_to_wait(&rl->drain, &wait, TASK_UNINTERRUPTIBLE);
-
- if (wait_drain(q, rl, wait_dispatch)) {
- __generic_unplug_device(q);
- spin_unlock_irq(q->queue_lock);
- io_schedule();
- spin_lock_irq(q->queue_lock);
- }
-
- finish_wait(&rl->drain, &wait);
- }
-
- spin_unlock_irq(q->queue_lock);
-}
-
-/*
- * block waiting for the io scheduler being started again.
- */
-static inline void block_wait_queue_running(request_queue_t *q)
-{
- DEFINE_WAIT(wait);
-
- while (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) {
- struct request_list *rl = &q->rq;
-
- prepare_to_wait_exclusive(&rl->drain, &wait,
- TASK_UNINTERRUPTIBLE);
-
- /*
- * re-check the condition. avoids using prepare_to_wait()
- * in the fast path (queue is running)
- */
- if (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))
- io_schedule();
-
- finish_wait(&rl->drain, &wait);
- }
-}
-
static void handle_bad_sector(struct bio *bio)
{
char b[BDEVNAME_SIZE];
@@ -2988,8 +2902,6 @@ end_io:
if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
goto end_io;
- block_wait_queue_running(q);
-
/*
* If this device has partitions, remap block n
* of partition p to block n+start(p) of the disk.