public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Thomas Gleixner <tglx@linutronix.de>
To: LKML <linux-kernel@vger.kernel.org>
Cc: Zhen Lei <thunder.leizhen@huawei.com>, Waiman Long <longman@redhat.com>
Subject: [patch 21/25] debugobjects: Implement batch processing
Date: Mon,  7 Oct 2024 18:50:17 +0200 (CEST)	[thread overview]
Message-ID: <20241007164914.258995000@linutronix.de> (raw)
In-Reply-To: 20241007163507.647617031@linutronix.de

Adding and removing single objects in a loop is bad in terms of lock
contention and cache line accesses.

To implement batching, record the last object in a batch in the object
itself. This is trivialy possible as hlists are strictly stacks. At a batch
boundary, when the first object is added to the list the object stores a
pointer to itself in debug_obj::batch_last. When the next object is added
to the list then the batch_last pointer is retrieved from the first object
in the list and stored in the to be added one.

That means for batch processing the first object always has a pointer to
the last object in a batch, which allows to move batches in a cache line
efficient way and reduces the lock held time.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 lib/debugobjects.c |   61 +++++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 46 insertions(+), 15 deletions(-)

--- a/lib/debugobjects.c
+++ b/lib/debugobjects.c
@@ -149,18 +149,31 @@ static __always_inline bool pool_must_re
 
 static bool pool_move_batch(struct obj_pool *dst, struct obj_pool *src)
 {
-	if (dst->cnt + ODEBUG_BATCH_SIZE > dst->max_cnt || !src->cnt)
+	struct hlist_node *last, *next_batch, *first_batch;
+	struct debug_obj *obj;
+
+	if (dst->cnt >= dst->max_cnt || !src->cnt)
 		return false;
 
-	for (int i = 0; i < ODEBUG_BATCH_SIZE && src->cnt; i++) {
-		struct hlist_node *node = src->objects.first;
+	first_batch = src->objects.first;
+	obj = hlist_entry(first_batch, typeof(*obj), node);
+	last = obj->batch_last;
+	next_batch = last->next;
 
-		WRITE_ONCE(src->cnt, src->cnt - 1);
-		WRITE_ONCE(dst->cnt, dst->cnt + 1);
+	/* Move the next batch to the front of the source pool */
+	src->objects.first = next_batch;
+	if (next_batch)
+		next_batch->pprev = &src->objects.first;
+
+	/* Add the extracted batch to the destination pool */
+	last->next = dst->objects.first;
+	if (last->next)
+		last->next->pprev = &last->next;
+	first_batch->pprev = &dst->objects.first;
+	dst->objects.first = first_batch;
 
-		hlist_del(node);
-		hlist_add_head(node, &dst->objects);
-	}
+	WRITE_ONCE(src->cnt, src->cnt - ODEBUG_BATCH_SIZE);
+	WRITE_ONCE(dst->cnt, dst->cnt + ODEBUG_BATCH_SIZE);
 	return true;
 }
 
@@ -182,16 +195,27 @@ static bool pool_push_batch(struct obj_p
 
 static bool pool_pop_batch(struct hlist_head *head, struct obj_pool *src)
 {
+	struct hlist_node *last, *next;
+	struct debug_obj *obj;
+
 	if (!src->cnt)
 		return false;
 
-	for (int i = 0; src->cnt && i < ODEBUG_BATCH_SIZE; i++) {
-		struct hlist_node *node = src->objects.first;
+	/* Move the complete list to the head */
+	hlist_move_list(&src->objects, head);
 
-		WRITE_ONCE(src->cnt, src->cnt - 1);
-		hlist_del(node);
-		hlist_add_head(node, head);
-	}
+	obj = hlist_entry(head->first, typeof(*obj), node);
+	last = obj->batch_last;
+	next = last->next;
+	/* Disconnect the batch from the list */
+	last->next = NULL;
+
+	/* Move the node after last back to the source pool. */
+	src->objects.first = next;
+	if (next)
+		next->pprev = &src->objects.first;
+
+	WRITE_ONCE(src->cnt, src->cnt - ODEBUG_BATCH_SIZE);
 	return true;
 }
 
@@ -226,7 +250,7 @@ static struct debug_obj *pcpu_alloc(void
 			if (!pool_move_batch(pcp, &pool_global))
 				return NULL;
 		}
-		obj_pool_used += pcp->cnt;
+		obj_pool_used += ODEBUG_BATCH_SIZE;
 
 		if (obj_pool_used > obj_pool_max_used)
 			obj_pool_max_used = obj_pool_used;
@@ -239,9 +263,16 @@ static struct debug_obj *pcpu_alloc(void
 static void pcpu_free(struct debug_obj *obj)
 {
 	struct obj_pool *pcp = this_cpu_ptr(&pool_pcpu);
+	struct debug_obj *first;
 
 	lockdep_assert_irqs_disabled();
 
+	if (!(pcp->cnt % ODEBUG_BATCH_SIZE)) {
+		obj->batch_last = &obj->node;
+	} else {
+		first = hlist_entry(pcp->objects.first, typeof(*first), node);
+		obj->batch_last = first->batch_last;
+	}
 	hlist_add_head(&obj->node, &pcp->objects);
 	pcp->cnt++;
 


  parent reply	other threads:[~2024-10-07 16:50 UTC|newest]

Thread overview: 81+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-10-07 16:49 [patch 00/25] debugobjects: Rework object handling Thomas Gleixner
2024-10-07 16:49 ` [patch 01/25] debugobjects: Delete a piece of redundant code Thomas Gleixner
2024-10-07 16:49 ` [patch 02/25] debugobjects: Collect newly allocated objects in a list to reduce lock contention Thomas Gleixner
2024-10-07 16:49 ` [patch 03/25] debugobjects: Dont destroy kmem cache in init() Thomas Gleixner
2024-10-10  2:14   ` Leizhen (ThunderTown)
2024-10-10 11:46     ` Thomas Gleixner
2024-10-10 13:31       ` Leizhen (ThunderTown)
2024-10-11 20:37         ` Thomas Gleixner
2024-10-12  1:50           ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:49 ` [patch 04/25] debugobjects: Remove pointless hlist initialization Thomas Gleixner
2024-10-10  2:19   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:49 ` [patch 05/25] debugobjects: Dont free objects directly on CPU hotplug Thomas Gleixner
2024-10-10  2:33   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:49 ` [patch 06/25] debugobjects: Reuse put_objects() on OOM Thomas Gleixner
2024-10-10  2:38   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:49 ` [patch 07/25] debugobjects: Remove pointless debug printk Thomas Gleixner
2024-10-10  2:44   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 08/25] debugobjects: Provide and use free_object_list() Thomas Gleixner
2024-10-10  2:54   ` Leizhen (ThunderTown)
2024-10-11 20:40     ` Thomas Gleixner
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 09/25] debugobjects: Make debug_objects_enabled bool Thomas Gleixner
2024-10-10  3:00   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 10/25] debugobjects: Reduce parallel pool fill attempts Thomas Gleixner
2024-10-07 16:50 ` [patch 11/25] debugobjects: Move pools into a datastructure Thomas Gleixner
2024-10-10  3:47   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 12/25] debugobjects: Use separate list head for boot pool Thomas Gleixner
2024-10-10  4:04   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 13/25] debugobjects: Rename and tidy up per CPU pools Thomas Gleixner
2024-10-10  6:23   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 14/25] debugobjects: Move min/max count into pool struct Thomas Gleixner
2024-10-10  6:26   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 15/25] debugobjects: Rework object allocation Thomas Gleixner
2024-10-10  6:39   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 16/25] debugobjects: Rework object freeing Thomas Gleixner
2024-10-10  7:39   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 17/25] debugobjects: Rework free_object_work() Thomas Gleixner
2024-10-10  8:10   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 18/25] debugobjects: Use static key for boot pool selection Thomas Gleixner
2024-10-10  8:12   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 19/25] debugobjects: Prepare for batching Thomas Gleixner
2024-10-10  8:15   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 20/25] debugobjects: Prepare kmem_cache allocations " Thomas Gleixner
2024-10-10  8:40   ` Leizhen (ThunderTown)
2024-10-11 20:47     ` Thomas Gleixner
2024-10-12  2:02       ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` Thomas Gleixner [this message]
2024-10-10  9:39   ` [patch 21/25] debugobjects: Implement batch processing Leizhen (ThunderTown)
2024-10-11 20:48     ` Thomas Gleixner
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 22/25] debugobjects: Move pool statistics into global_pool struct Thomas Gleixner
2024-10-10  9:50   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 23/25] debugobjects: Double the per CPU slots Thomas Gleixner
2024-10-10  9:51   ` Leizhen (ThunderTown)
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 24/25] debugobjects: Refill per CPU pool more agressively Thomas Gleixner
2024-10-10 10:02   ` Leizhen (ThunderTown)
2024-10-11 20:49     ` Thomas Gleixner
2024-10-15 15:36   ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner
2024-10-07 16:50 ` [patch 25/25] debugobjects: Track object usage to avoid premature freeing of objects Thomas Gleixner
2024-10-10 13:13   ` Leizhen (ThunderTown)
2024-10-13 18:45     ` Thomas Gleixner
2024-10-14  1:46       ` Leizhen (ThunderTown)
2024-10-15 15:36       ` [tip: core/debugobjects] " tip-bot2 for Thomas Gleixner

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20241007164914.258995000@linutronix.de \
    --to=tglx@linutronix.de \
    --cc=linux-kernel@vger.kernel.org \
    --cc=longman@redhat.com \
    --cc=thunder.leizhen@huawei.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox