linux-kernel - [PATCHv2 2/2] blk-mq: add support for CPU latency limits

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20241018075416.436916-3-tero.kristo@linux.intel.com>
Date: Fri, 18 Oct 2024 10:30:38 +0300
From: Tero Kristo <tero.kristo@...ux.intel.com>
To: axboe@...nel.dk
Cc: hch@....de,
	linux-block@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: [PATCHv2 2/2] blk-mq: add support for CPU latency limits

Add support for setting CPU latency limits when a request is dispatched
to driver layer, and removing it once the device is idle. The latency
limits use the dev PM QoS framework for setting per-cpu limits for
active CPUs. The feature is user configurable via sysfs knobs under the
block device.

Signed-off-by: Tero Kristo <tero.kristo@...ux.intel.com>
---
 block/blk-mq.c         | 54 ++++++++++++++++++++++++++++++++++++++++++
 include/linux/blk-mq.h | 12 ++++++++++
 2 files changed, 66 insertions(+)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 4b2c8e940f59..f8906e2aff6d 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -29,6 +29,7 @@
 #include <linux/blk-crypto.h>
 #include <linux/part_stat.h>
 #include <linux/sched/isolation.h>
+#include <linux/pm_qos.h>
 
 #include <trace/events/block.h>
 
@@ -2700,11 +2701,62 @@ static void blk_mq_plug_issue_direct(struct blk_plug *plug)
 static void __blk_mq_flush_plug_list(struct request_queue *q,
 				     struct blk_plug *plug)
 {
+	struct request *req, *next;
+	struct blk_mq_hw_ctx *hctx;
+	int cpu;
+
 	if (blk_queue_quiesced(q))
 		return;
+
+	rq_list_for_each_safe(&plug->mq_list, req, next) {
+		hctx = req->mq_hctx;
+
+		if (next && next->mq_hctx == hctx)
+			continue;
+
+		if (q->disk->cpu_lat_limit < 0)
+			continue;
+
+		hctx->last_active = jiffies + msecs_to_jiffies(q->disk->cpu_lat_timeout);
+
+		if (!hctx->cpu_lat_limit_active) {
+			hctx->cpu_lat_limit_active = true;
+			for_each_cpu(cpu, hctx->cpumask) {
+				struct dev_pm_qos_request *qos;
+
+				qos = per_cpu_ptr(hctx->cpu_lat_qos, cpu);
+				dev_pm_qos_add_request(get_cpu_device(cpu), qos,
+						       DEV_PM_QOS_RESUME_LATENCY,
+						       q->disk->cpu_lat_limit);
+			}
+			schedule_delayed_work(&hctx->cpu_latency_work,
+					      msecs_to_jiffies(q->disk->cpu_lat_timeout));
+		}
+	}
+
 	q->mq_ops->queue_rqs(&plug->mq_list);
 }
 
+static void blk_mq_cpu_latency_work(struct work_struct *work)
+{
+	struct blk_mq_hw_ctx *hctx = container_of(work, struct blk_mq_hw_ctx,
+						  cpu_latency_work.work);
+	int cpu;
+
+	if (time_after(jiffies, hctx->last_active)) {
+		for_each_cpu(cpu, hctx->cpumask) {
+			struct dev_pm_qos_request *qos;
+
+			qos = per_cpu_ptr(hctx->cpu_lat_qos, cpu);
+			dev_pm_qos_remove_request(qos);
+		}
+		hctx->cpu_lat_limit_active = false;
+	} else {
+		schedule_delayed_work(&hctx->cpu_latency_work,
+				      msecs_to_jiffies(hctx->queue->disk->cpu_lat_timeout));
+	}
+}
+
 static void blk_mq_dispatch_plug_list(struct blk_plug *plug, bool from_sched)
 {
 	struct blk_mq_hw_ctx *this_hctx = NULL;
@@ -3729,6 +3778,11 @@ static int blk_mq_init_hctx(struct request_queue *q,
 	if (xa_insert(&q->hctx_table, hctx_idx, hctx, GFP_KERNEL))
 		goto exit_flush_rq;
 
+	hctx->cpu_lat_qos = alloc_percpu(struct dev_pm_qos_request);
+	if (!hctx->cpu_lat_qos)
+		goto exit_flush_rq;
+	INIT_DELAYED_WORK(&hctx->cpu_latency_work, blk_mq_cpu_latency_work);
+
 	return 0;
 
  exit_flush_rq:
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index b751cc92209b..2b61942490d6 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -435,6 +435,18 @@ struct blk_mq_hw_ctx {
 	/** @kobj: Kernel object for sysfs. */
 	struct kobject		kobj;
 
+	/** @cpu_latency_work: Work to handle CPU latency PM limits. */
+	struct delayed_work	cpu_latency_work;
+
+	/** @cpu_lat_limit_active: If CPU latency limits are active or not. */
+	bool			cpu_lat_limit_active;
+
+	/** @last_active: Jiffies value when the queue was last active. */
+	unsigned long		last_active;
+
+	/** @cpu_lat_qos: PM QoS latency limits for individual CPUs. */
+	struct dev_pm_qos_request __percpu *cpu_lat_qos;
+
 #ifdef CONFIG_BLK_DEBUG_FS
 	/**
 	 * @debugfs_dir: debugfs directory for this hardware queue. Named
-- 
2.43.1