From: Liu Xiang <liu.xiang@zlingsmart.com>
To: linux-block@vger.kernel.org
Cc: axboe@kernel.dk, linux-kernel@vger.kernel.org,
Liu Xiang <liu.xiang@zlingsmart.com>
Subject: [PATCH] blk-mq: introduce REQ_COMPLETE_WQ and add a workqueue to complete the request
Date: Wed, 20 Jan 2021 10:15:22 +0800 [thread overview]
Message-ID: <20210120021522.28584-1-liu.xiang@zlingsmart.com> (raw)
The commit "40d09b53bfc557af7481b9d80f060a7ac9c7d314" has solved the
irqsoff problem by completing the request in softirq. But it may cause
the system to suffer bad preemptoff time.
Introduce the REQ_COMPLETE_WQ flag and blk_complete workqueue.
This flag makes the request to be completed in the blk_complete workqueue.
It can be used for requests that want to cut down both irqsoff and
preemptoff time.
Signed-off-by: Liu Xiang <liu.xiang@zlingsmart.com>
---
block/blk-mq.c | 46 +++++++++++++++++++++++++++++++++++++++
include/linux/blk_types.h | 4 ++++
include/linux/blkdev.h | 5 +++++
3 files changed, 55 insertions(+)
diff --git a/block/blk-mq.c b/block/blk-mq.c
index f285a9123..c707582ba 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -42,6 +42,10 @@
#include "blk-rq-qos.h"
static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
+/*
+ * Controlling structure for block complete
+ */
+static struct workqueue_struct *blk_complete_wq;
static void blk_mq_poll_stats_start(struct request_queue *q);
static void blk_mq_poll_stats_fn(struct blk_stat_callback *cb);
@@ -567,6 +571,26 @@ void blk_mq_end_request(struct request *rq, blk_status_t error)
}
EXPORT_SYMBOL(blk_mq_end_request);
+static void blk_mq_complete_work(struct work_struct *work)
+{
+ struct request_queue *q =
+ container_of(work, struct request_queue, complete_work);
+ struct list_head local_list;
+ unsigned long flags;
+
+ spin_lock_irqsave(&q->complete_lock, flags);
+ list_replace_init(&q->complete_list, &local_list);
+ spin_unlock_irqrestore(&q->complete_lock, flags);
+
+ while (!list_empty(&local_list)) {
+ struct request *rq;
+
+ rq = list_entry(local_list.next, struct request, complete_list);
+ list_del_init(&rq->complete_list);
+ rq->q->mq_ops->complete(rq);
+ }
+}
+
/*
* Softirq action handler - move entries to local list and loop over them
* while passing them to the queue registered handler.
@@ -680,6 +704,19 @@ bool blk_mq_complete_request_remote(struct request *rq)
if (rq->cmd_flags & REQ_HIPRI)
return false;
+ /*
+ * For a request that wants to complete in workqueue,
+ */
+ if (rq->cmd_flags & REQ_COMPLETE_WQ) {
+ unsigned long flags;
+
+ spin_lock_irqsave(&rq->q->complete_lock, flags);
+ list_add_tail(&rq->complete_list, &rq->q->complete_list);
+ spin_unlock_irqrestore(&rq->q->complete_lock, flags);
+ queue_work(blk_complete_wq, &rq->q->complete_work);
+ return true;
+ }
+
if (blk_mq_complete_need_ipi(rq)) {
INIT_CSD(&rq->csd, __blk_mq_complete_request_remote, rq);
smp_call_function_single_async(rq->mq_ctx->cpu, &rq->csd);
@@ -3211,6 +3248,10 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
INIT_LIST_HEAD(&q->requeue_list);
spin_lock_init(&q->requeue_lock);
+ INIT_WORK(&q->complete_work, blk_mq_complete_work);
+ INIT_LIST_HEAD(&q->complete_list);
+ spin_lock_init(&q->complete_lock);
+
q->nr_requests = set->queue_depth;
/*
@@ -3907,6 +3948,11 @@ static int __init blk_mq_init(void)
INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
open_softirq(BLOCK_SOFTIRQ, blk_done_softirq);
+ blk_complete_wq = alloc_workqueue("blk_complete",
+ WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
+ if (!blk_complete_wq)
+ panic("Failed to create blk_complete\n");
+
cpuhp_setup_state_nocalls(CPUHP_BLOCK_SOFTIRQ_DEAD,
"block/softirq:dead", NULL,
blk_softirq_cpu_dead);
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 866f74261..251110fd9 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -421,6 +421,9 @@ enum req_flag_bits {
__REQ_HIPRI,
+ /* do req complete in workqueue */
+ __REQ_COMPLETE_WQ,
+
/* for driver use */
__REQ_DRV,
__REQ_SWAP, /* swapping request. */
@@ -445,6 +448,7 @@ enum req_flag_bits {
#define REQ_NOUNMAP (1ULL << __REQ_NOUNMAP)
#define REQ_HIPRI (1ULL << __REQ_HIPRI)
+#define REQ_COMPLETE_WQ (1ULL << __REQ_COMPLETE_WQ)
#define REQ_DRV (1ULL << __REQ_DRV)
#define REQ_SWAP (1ULL << __REQ_SWAP)
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index f94ee3089..758aff8f0 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -143,6 +143,7 @@ struct request {
struct bio *biotail;
struct list_head queuelist;
+ struct list_head complete_list;
/*
* The hash is used inside the scheduler, and killed once the
@@ -543,6 +544,10 @@ struct request_queue {
spinlock_t requeue_lock;
struct delayed_work requeue_work;
+ struct list_head complete_list;
+ spinlock_t complete_lock;
+ struct work_struct complete_work;
+
struct mutex sysfs_lock;
struct mutex sysfs_dir_lock;
--
2.17.1
next reply other threads:[~2021-01-20 2:17 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-20 2:15 Liu Xiang [this message]
2021-01-20 9:54 ` [PATCH] blk-mq: introduce REQ_COMPLETE_WQ and add a workqueue to complete the request Christoph Hellwig
2021-01-21 7:33 ` liu xiang
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=20210120021522.28584-1-liu.xiang@zlingsmart.com \
--to=liu.xiang@zlingsmart.com \
--cc=axboe@kernel.dk \
--cc=linux-block@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
/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;
as well as URLs for NNTP newsgroup(s).