block: make unplug timer trace event correspond to the schedule() unplug

It's a pretty close match to what we had before - the timer triggering would mean that nobody unplugged the plug in due time, in the new scheme this matches very closely what the schedule() unplug now is. It's essentially the difference between an explicit unplug (IO unplug) or an implicit unplug (timer unplug, we scheduled with pending IO queued). Signed-off-by: Jens Axboe <jaxboe@fusionio.com>

block: make unplug timer trace event correspond to the schedule() unplug
It's a pretty close match to what we had before - the timer triggering would mean that nobody unplugged the plug in due time, in the new scheme this matches very closely what the schedule() unplug now is. It's essentially the difference between an explicit unplug (IO unplug) or an implicit unplug (timer unplug, we scheduled with pending IO queued). Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Jens Axboe
1 parent a237c1c5bc
Showing 3 changed files with 31 additions and 18 deletions Side-by-side Diff
block/blk-core.c
include/trace/events/block.h
kernel/trace/blktrace.c
@@ -2662,17 +2662,23 @@
 	return !(rqa->q <= rqb->q);
 }
  
+/*
+ * If 'from_schedule' is true, then postpone the dispatch of requests
+ * until a safe kblockd context. We due this to avoid accidental big
+ * additional stack usage in driver dispatch, in places where the originally
+ * plugger did not intend it.
+ */
 static void queue_unplugged(struct request_queue *q, unsigned int depth,
-			    bool force_kblockd)
+			    bool from_schedule)
 {
-	trace_block_unplug_io(q, depth);
-	__blk_run_queue(q, force_kblockd);
+	trace_block_unplug(q, depth, !from_schedule);
+	__blk_run_queue(q, from_schedule);
  
 	if (q->unplugged_fn)
 		q->unplugged_fn(q);
 }
  
-void blk_flush_plug_list(struct blk_plug *plug, bool force_kblockd)
+void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
 {
 	struct request_queue *q;
 	unsigned long flags;
@@ -2707,7 +2713,7 @@
 		BUG_ON(!rq->q);
 		if (rq->q != q) {
 			if (q) {
-				queue_unplugged(q, depth, force_kblockd);
+				queue_unplugged(q, depth, from_schedule);
 				spin_unlock(q->queue_lock);
 			}
 			q = rq->q;
@@ -2728,7 +2734,7 @@
 	}
  
 	if (q) {
-		queue_unplugged(q, depth, force_kblockd);
+		queue_unplugged(q, depth, from_schedule);
 		spin_unlock(q->queue_lock);
 	}
  
@@ -401,9 +401,9 @@
  
 DECLARE_EVENT_CLASS(block_unplug,
  
-	TP_PROTO(struct request_queue *q, unsigned int depth),
+	TP_PROTO(struct request_queue *q, unsigned int depth, bool explicit),
  
-	TP_ARGS(q, depth),
+	TP_ARGS(q, depth, explicit),
  
 	TP_STRUCT__entry(
 		__field( int,		nr_rq			)
  
  
  
  
@@ -419,18 +419,19 @@
 );
  
 /**
- * block_unplug_io - release of operations requests in request queue
+ * block_unplug - release of operations requests in request queue
  * @q: request queue to unplug
  * @depth: number of requests just added to the queue
+ * @explicit: whether this was an explicit unplug, or one from schedule()
  *
  * Unplug request queue @q because device driver is scheduled to work
  * on elements in the request queue.
  */
-DEFINE_EVENT(block_unplug, block_unplug_io,
+DEFINE_EVENT(block_unplug, block_unplug,
  
-	TP_PROTO(struct request_queue *q, unsigned int depth),
+	TP_PROTO(struct request_queue *q, unsigned int depth, bool explicit),
  
-	TP_ARGS(q, depth)
+	TP_ARGS(q, depth, explicit)
 );
  
 /**
@@ -850,16 +850,21 @@
 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 }
  
-static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q,
-				    unsigned int depth)
+static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
+				    unsigned int depth, bool explicit)
 {
 	struct blk_trace *bt = q->blk_trace;
  
 	if (bt) {
 		__be64 rpdu = cpu_to_be64(depth);
+		u32 what;
  
-		__blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
-				sizeof(rpdu), &rpdu);
+		if (explicit)
+			what = BLK_TA_UNPLUG_IO;
+		else
+			what = BLK_TA_UNPLUG_TIMER;
+
+		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 	}
 }
  
@@ -1002,7 +1007,7 @@
 	WARN_ON(ret);
 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
 	WARN_ON(ret);
-	ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
+	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
 	WARN_ON(ret);
 	ret = register_trace_block_split(blk_add_trace_split, NULL);
 	WARN_ON(ret);
@@ -1017,7 +1022,7 @@
 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
 	unregister_trace_block_split(blk_add_trace_split, NULL);
-	unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
+	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
@@ -1332,6 +1337,7 @@
 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
+	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
...	...	@@ -2662,17 +2662,23 @@
2662	2662	return !(rqa->q <= rqb->q);
2663	2663	}
2664	2664
	2665	+/*
	2666	+ * If 'from_schedule' is true, then postpone the dispatch of requests
	2667	+ * until a safe kblockd context. We due this to avoid accidental big
	2668	+ * additional stack usage in driver dispatch, in places where the originally
	2669	+ * plugger did not intend it.
	2670	+ */
2665	2671	static void queue_unplugged(struct request_queue *q, unsigned int depth,
2666		- bool force_kblockd)
	2672	+ bool from_schedule)
2667	2673	{
2668		- trace_block_unplug_io(q, depth);
2669		- __blk_run_queue(q, force_kblockd);
	2674	+ trace_block_unplug(q, depth, !from_schedule);
	2675	+ __blk_run_queue(q, from_schedule);
2670	2676
2671	2677	if (q->unplugged_fn)
2672	2678	q->unplugged_fn(q);
2673	2679	}
2674	2680
2675		-void blk_flush_plug_list(struct blk_plug *plug, bool force_kblockd)
	2681	+void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2676	2682	{
2677	2683	struct request_queue *q;
2678	2684	unsigned long flags;
...	...	@@ -2707,7 +2713,7 @@
2707	2713	BUG_ON(!rq->q);
2708	2714	if (rq->q != q) {
2709	2715	if (q) {
2710		- queue_unplugged(q, depth, force_kblockd);
	2716	+ queue_unplugged(q, depth, from_schedule);
2711	2717	spin_unlock(q->queue_lock);
2712	2718	}
2713	2719	q = rq->q;
...	...	@@ -2728,7 +2734,7 @@
2728	2734	}
2729	2735
2730	2736	if (q) {
2731		- queue_unplugged(q, depth, force_kblockd);
	2737	+ queue_unplugged(q, depth, from_schedule);
2732	2738	spin_unlock(q->queue_lock);
2733	2739	}
2734	2740
...	...	@@ -401,9 +401,9 @@
401	401
402	402	DECLARE_EVENT_CLASS(block_unplug,
403	403
404		- TP_PROTO(struct request_queue *q, unsigned int depth),
	404	+ TP_PROTO(struct request_queue *q, unsigned int depth, bool explicit),
405	405
406		- TP_ARGS(q, depth),
	406	+ TP_ARGS(q, depth, explicit),
407	407
408	408	TP_STRUCT__entry(
409	409	__field( int, nr_rq )
410	410
411	411
412	412
413	413
...	...	@@ -419,18 +419,19 @@
419	419	);
420	420
421	421	/**
422		- * block_unplug_io - release of operations requests in request queue
	422	+ * block_unplug - release of operations requests in request queue
423	423	* @q: request queue to unplug
424	424	* @depth: number of requests just added to the queue
	425	+ * @explicit: whether this was an explicit unplug, or one from schedule()
425	426	*
426	427	* Unplug request queue @q because device driver is scheduled to work
427	428	* on elements in the request queue.
428	429	*/
429		-DEFINE_EVENT(block_unplug, block_unplug_io,
	430	+DEFINE_EVENT(block_unplug, block_unplug,
430	431
431		- TP_PROTO(struct request_queue *q, unsigned int depth),
	432	+ TP_PROTO(struct request_queue *q, unsigned int depth, bool explicit),
432	433
433		- TP_ARGS(q, depth)
	434	+ TP_ARGS(q, depth, explicit)
434	435	);
435	436
436	437	/**
...	...	@@ -850,16 +850,21 @@
850	850	__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
851	851	}
852	852
853		-static void blk_add_trace_unplug_io(void ignore, struct request_queue q,
854		- unsigned int depth)
	853	+static void blk_add_trace_unplug(void ignore, struct request_queue q,
	854	+ unsigned int depth, bool explicit)
855	855	{
856	856	struct blk_trace *bt = q->blk_trace;
857	857
858	858	if (bt) {
859	859	__be64 rpdu = cpu_to_be64(depth);
	860	+ u32 what;
860	861
861		- __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
862		- sizeof(rpdu), &rpdu);
	862	+ if (explicit)
	863	+ what = BLK_TA_UNPLUG_IO;
	864	+ else
	865	+ what = BLK_TA_UNPLUG_TIMER;
	866	+
	867	+ __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
863	868	}
864	869	}
865	870
...	...	@@ -1002,7 +1007,7 @@
1002	1007	WARN_ON(ret);
1003	1008	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1004	1009	WARN_ON(ret);
1005		- ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
	1010	+ ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1006	1011	WARN_ON(ret);
1007	1012	ret = register_trace_block_split(blk_add_trace_split, NULL);
1008	1013	WARN_ON(ret);
...	...	@@ -1017,7 +1022,7 @@
1017	1022	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1018	1023	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1019	1024	unregister_trace_block_split(blk_add_trace_split, NULL);
1020		- unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
	1025	+ unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1021	1026	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1022	1027	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1023	1028	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
...	...	@@ -1332,6 +1337,7 @@
1332	1337	[__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1333	1338	[__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1334	1339	[__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
	1340	+ [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1335	1341	[__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1336	1342	[__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1337	1343	[__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },