[PATCH 1/3] block: Implement a blk_yield function to voluntarily give up the I/O scheduler.
From: Jeff Moyer
Date: Tue Jun 22 2010 - 17:36:10 EST
This patch implements a blk_yield to allow a process to voluntarily
give up its I/O scheduler time slice. This is desirable for those processes
which know that they will be blocked on I/O from another process, such as
the file system journal thread. Following patches will put calls to blk_yield
into jbd and jbd2.
Signed-off-by: Jeff Moyer <jmoyer@xxxxxxxxxx>
---
block/blk-core.c | 13 +++++
block/blk-settings.c | 6 ++
block/cfq-iosched.c | 123 +++++++++++++++++++++++++++++++++++++++++++++-
block/elevator.c | 8 +++
include/linux/blkdev.h | 4 ++
include/linux/elevator.h | 3 +
6 files changed, 155 insertions(+), 2 deletions(-)
diff --git a/block/blk-core.c b/block/blk-core.c
index f84cce4..b9afbba 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -324,6 +324,18 @@ void blk_unplug(struct request_queue *q)
}
EXPORT_SYMBOL(blk_unplug);
+void generic_yield_iosched(struct request_queue *q, struct task_struct *tsk)
+{
+ elv_yield(q, tsk);
+}
+
+void blk_yield(struct request_queue *q, struct task_struct *tsk)
+{
+ if (q->yield_fn)
+ q->yield_fn(q, tsk);
+}
+EXPORT_SYMBOL(blk_yield);
+
/**
* blk_start_queue - restart a previously stopped queue
* @q: The &struct request_queue in question
@@ -609,6 +621,7 @@ blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
q->request_fn = rfn;
q->prep_rq_fn = NULL;
q->unplug_fn = generic_unplug_device;
+ q->yield_fn = generic_yield_iosched;
q->queue_flags = QUEUE_FLAG_DEFAULT;
q->queue_lock = lock;
diff --git a/block/blk-settings.c b/block/blk-settings.c
index f5ed5a1..fe548c9 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -171,6 +171,12 @@ void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
}
EXPORT_SYMBOL(blk_queue_make_request);
+void blk_queue_yield(struct request_queue *q, yield_fn *yield)
+{
+ q->yield_fn = yield;
+}
+EXPORT_SYMBOL_GPL(blk_queue_yield);
+
/**
* blk_queue_bounce_limit - set bounce buffer limit for queue
* @q: the request queue for the device
diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index dab836e..a9922b9 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -87,9 +87,12 @@ struct cfq_rb_root {
unsigned total_weight;
u64 min_vdisktime;
struct rb_node *active;
+ unsigned long last_expiry;
+ pid_t last_pid;
};
#define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \
- .count = 0, .min_vdisktime = 0, }
+ .count = 0, .min_vdisktime = 0, .last_expiry = 0UL, \
+ .last_pid = (pid_t)-1, }
/*
* Per process-grouping structure
@@ -147,6 +150,7 @@ struct cfq_queue {
struct cfq_queue *new_cfqq;
struct cfq_group *cfqg;
struct cfq_group *orig_cfqg;
+ struct cfq_io_context *yield_to;
};
/*
@@ -318,6 +322,7 @@ enum cfqq_state_flags {
CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */
CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */
CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */
+ CFQ_CFQQ_FLAG_yield, /* Allow another cfqq to run */
};
#define CFQ_CFQQ_FNS(name) \
@@ -347,6 +352,7 @@ CFQ_CFQQ_FNS(coop);
CFQ_CFQQ_FNS(split_coop);
CFQ_CFQQ_FNS(deep);
CFQ_CFQQ_FNS(wait_busy);
+CFQ_CFQQ_FNS(yield);
#undef CFQ_CFQQ_FNS
#ifdef CONFIG_CFQ_GROUP_IOSCHED
@@ -1614,6 +1620,15 @@ __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
cfq_clear_cfqq_wait_request(cfqq);
cfq_clear_cfqq_wait_busy(cfqq);
+ if (!cfq_cfqq_yield(cfqq)) {
+ struct cfq_rb_root *st;
+ st = service_tree_for(cfqq->cfqg,
+ cfqq_prio(cfqq), cfqq_type(cfqq));
+ st->last_expiry = jiffies;
+ st->last_pid = cfqq->pid;
+ }
+ cfq_clear_cfqq_yield(cfqq);
+
/*
* If this cfqq is shared between multiple processes, check to
* make sure that those processes are still issuing I/Os within
@@ -2118,7 +2133,7 @@ static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg)
slice = max(slice, 2 * cfqd->cfq_slice_idle);
slice = max_t(unsigned, slice, CFQ_MIN_TT);
- cfq_log(cfqd, "workload slice:%d", slice);
+ cfq_log(cfqd, "workload:%d slice:%d", cfqd->serving_type, slice);
cfqd->workload_expires = jiffies + slice;
cfqd->noidle_tree_requires_idle = false;
}
@@ -2153,6 +2168,36 @@ static void cfq_choose_cfqg(struct cfq_data *cfqd)
choose_service_tree(cfqd, cfqg);
}
+static int cfq_should_yield_now(struct cfq_queue *cfqq,
+ struct cfq_queue **yield_to)
+{
+ struct cfq_queue *new_cfqq;
+
+ new_cfqq = cic_to_cfqq(cfqq->yield_to, 1);
+
+ /*
+ * If the queue we're yielding to is in a different cgroup,
+ * just expire our own time slice.
+ */
+ if (new_cfqq->cfqg != cfqq->cfqg) {
+ *yield_to = NULL;
+ return 1;
+ }
+
+ /*
+ * If the new queue has pending I/O, then switch to it
+ * immediately. Otherwise, see if we can idle until it is
+ * ready to preempt us.
+ */
+ if (!RB_EMPTY_ROOT(&new_cfqq->sort_list)) {
+ *yield_to = new_cfqq;
+ return 1;
+ }
+
+ *yield_to = NULL;
+ return 0;
+}
+
/*
* Select a queue for service. If we have a current active queue,
* check whether to continue servicing it, or retrieve and set a new one.
@@ -2187,6 +2232,10 @@ static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
* have been idling all along on this queue and it should be
* ok to wait for this request to complete.
*/
+ if (cfq_cfqq_yield(cfqq) &&
+ cfq_should_yield_now(cfqq, &new_cfqq))
+ goto expire;
+
if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
&& cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
cfqq = NULL;
@@ -2215,6 +2264,9 @@ static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
goto expire;
}
+ if (cfq_cfqq_yield(cfqq) && cfq_should_yield_now(cfqq, &new_cfqq))
+ goto expire;
+
/*
* No requests pending. If the active queue still has requests in
* flight or is idling for a new request, allow either of these
@@ -2241,6 +2293,65 @@ keep_queue:
return cfqq;
}
+static inline int expiry_data_valid(struct cfq_rb_root *service_tree)
+{
+ return (service_tree->last_pid != (pid_t)-1 &&
+ service_tree->last_expiry != 0UL);
+}
+
+static void cfq_yield(struct request_queue *q, struct task_struct *tsk)
+{
+ struct cfq_data *cfqd = q->elevator->elevator_data;
+ struct cfq_io_context *cic, *new_cic;
+ struct cfq_queue *cfqq;
+
+ cic = cfq_cic_lookup(cfqd, current->io_context);
+ if (!cic)
+ return;
+
+ task_lock(tsk);
+ new_cic = cfq_cic_lookup(cfqd, tsk->io_context);
+ atomic_long_inc(&tsk->io_context->refcount);
+ task_unlock(tsk);
+ if (!new_cic)
+ goto out_dec;
+
+ spin_lock_irq(q->queue_lock);
+
+ cfqq = cic_to_cfqq(cic, 1);
+ if (!cfqq)
+ goto out_unlock;
+
+ /*
+ * If we are currently servicing the SYNC_NOIDLE_WORKLOAD, and we
+ * are idling on the last queue in that workload, *and* there are no
+ * potential dependent readers running currently, then go ahead and
+ * yield the queue.
+ */
+ if (cfqd->active_queue == cfqq &&
+ cfqd->serving_type == SYNC_NOIDLE_WORKLOAD) {
+ /*
+ * If there's been no I/O from another process in the idle
+ * slice time, then there is by definition no dependent
+ * read going on for this service tree.
+ */
+ if (expiry_data_valid(cfqq->service_tree) &&
+ time_before(cfqq->service_tree->last_expiry +
+ cfq_slice_idle, jiffies) &&
+ cfqq->service_tree->last_pid != cfqq->pid)
+ goto out_unlock;
+ }
+
+ cfq_log_cfqq(cfqd, cfqq, "yielding queue to %d", tsk->pid);
+ cfqq->yield_to = new_cic;
+ cfq_mark_cfqq_yield(cfqq);
+
+out_unlock:
+ spin_unlock_irq(q->queue_lock);
+out_dec:
+ put_io_context(tsk->io_context);
+}
+
static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
{
int dispatched = 0;
@@ -3123,6 +3234,13 @@ cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
if (!cfqq)
return false;
+ /*
+ * If the active queue yielded its timeslice to this queue, let
+ * it preempt.
+ */
+ if (cfq_cfqq_yield(cfqq) && RQ_CIC(rq) == cfqq->yield_to)
+ return true;
+
if (cfq_class_idle(new_cfqq))
return false;
@@ -3973,6 +4091,7 @@ static struct elevator_type iosched_cfq = {
.elevator_deactivate_req_fn = cfq_deactivate_request,
.elevator_queue_empty_fn = cfq_queue_empty,
.elevator_completed_req_fn = cfq_completed_request,
+ .elevator_yield_fn = cfq_yield,
.elevator_former_req_fn = elv_rb_former_request,
.elevator_latter_req_fn = elv_rb_latter_request,
.elevator_set_req_fn = cfq_set_request,
diff --git a/block/elevator.c b/block/elevator.c
index 923a913..5e33297 100644
--- a/block/elevator.c
+++ b/block/elevator.c
@@ -866,6 +866,14 @@ void elv_completed_request(struct request_queue *q, struct request *rq)
}
}
+void elv_yield(struct request_queue *q, struct task_struct *tsk)
+{
+ struct elevator_queue *e = q->elevator;
+
+ if (e && e->ops->elevator_yield_fn)
+ e->ops->elevator_yield_fn(q, tsk);
+}
+
#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
static ssize_t
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 09a8402..8d073c0 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -263,6 +263,7 @@ struct request_pm_state
typedef void (request_fn_proc) (struct request_queue *q);
typedef int (make_request_fn) (struct request_queue *q, struct bio *bio);
+typedef void (yield_fn) (struct request_queue *q, struct task_struct *tsk);
typedef int (prep_rq_fn) (struct request_queue *, struct request *);
typedef void (unplug_fn) (struct request_queue *);
@@ -345,6 +346,7 @@ struct request_queue
request_fn_proc *request_fn;
make_request_fn *make_request_fn;
+ yield_fn *yield_fn;
prep_rq_fn *prep_rq_fn;
unplug_fn *unplug_fn;
merge_bvec_fn *merge_bvec_fn;
@@ -837,6 +839,7 @@ extern int blk_execute_rq(struct request_queue *, struct gendisk *,
extern void blk_execute_rq_nowait(struct request_queue *, struct gendisk *,
struct request *, int, rq_end_io_fn *);
extern void blk_unplug(struct request_queue *q);
+extern void blk_yield(struct request_queue *q, struct task_struct *tsk);
static inline struct request_queue *bdev_get_queue(struct block_device *bdev)
{
@@ -929,6 +932,7 @@ extern struct request_queue *blk_init_allocated_queue(struct request_queue *,
request_fn_proc *, spinlock_t *);
extern void blk_cleanup_queue(struct request_queue *);
extern void blk_queue_make_request(struct request_queue *, make_request_fn *);
+extern void blk_queue_yield(struct request_queue *, yield_fn *);
extern void blk_queue_bounce_limit(struct request_queue *, u64);
extern void blk_queue_max_hw_sectors(struct request_queue *, unsigned int);
extern void blk_queue_max_segments(struct request_queue *, unsigned short);
diff --git a/include/linux/elevator.h b/include/linux/elevator.h
index 2c958f4..a68b5b1 100644
--- a/include/linux/elevator.h
+++ b/include/linux/elevator.h
@@ -23,6 +23,7 @@ typedef void (elevator_add_req_fn) (struct request_queue *, struct request *);
typedef int (elevator_queue_empty_fn) (struct request_queue *);
typedef struct request *(elevator_request_list_fn) (struct request_queue *, struct request *);
typedef void (elevator_completed_req_fn) (struct request_queue *, struct request *);
+typedef void (elevator_yield_fn) (struct request_queue *, struct task_struct *tsk);
typedef int (elevator_may_queue_fn) (struct request_queue *, int);
typedef int (elevator_set_req_fn) (struct request_queue *, struct request *, gfp_t);
@@ -48,6 +49,7 @@ struct elevator_ops
elevator_queue_empty_fn *elevator_queue_empty_fn;
elevator_completed_req_fn *elevator_completed_req_fn;
+ elevator_yield_fn *elevator_yield_fn;
elevator_request_list_fn *elevator_former_req_fn;
elevator_request_list_fn *elevator_latter_req_fn;
@@ -111,6 +113,7 @@ extern void elv_bio_merged(struct request_queue *q, struct request *,
struct bio *);
extern void elv_requeue_request(struct request_queue *, struct request *);
extern int elv_queue_empty(struct request_queue *);
+extern void elv_yield(struct request_queue *, struct task_struct *);
extern struct request *elv_former_request(struct request_queue *, struct request *);
extern struct request *elv_latter_request(struct request_queue *, struct request *);
extern int elv_register_queue(struct request_queue *q);
--
1.6.5.2
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/