[PATCH V1 0/4] Add per-cpu PM QoS support for QCOM UFS

[PATCH V1 0/4] Add per-cpu PM QoS support for QCOM UFS

[Maramaina Naresh]

Add per-cpu PM QoS support for ufs. This improves random io performance
by 20% for ufs.

tiotest benchmark tool io performance results on sm8550 platform:

1. Without PM QoS support
	Type (Speed in)    | Average of 6 iterations
	Random Write(IPOS) | 32201
	Random Read(IPOS)  | 32201

2. With PM QoS support
	Type (Speed in)    | Average of 6 iterations
	Random Write(IPOS) | 40833.5
	Random Read(IPOS)  | 40833.5
(Improvement % with PM QoS = ~20%).

Maramaina Naresh (4):
  dt-bindings: ufs: qcom: Add qos property
  ufs: ufs-qcom: Add per-cpu PM QoS support for ufs
  ufs: ufs-qcom: Add per-cpu PM QoS vote support for ufs
  arm64: dts: qcom: sm8550: Add per-cpu PM QoS support for ufs

 .../devicetree/bindings/ufs/qcom,ufs.yaml          |  16 +
 arch/arm64/boot/dts/qcom/sm8550.dtsi               |   9 +
 drivers/ufs/host/ufs-qcom.c                        | 340 ++++++++++++++++++++-
 drivers/ufs/host/ufs-qcom.h                        |  37 +++
 4 files changed, 401 insertions(+), 1 deletion(-)

-- 
2.7.4

[PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property

[Maramaina Naresh]

Add bindings for per-cpu QoS for QCOM UFS. This improves random io
performance by 20% for QCOM UFS.

Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
---
 Documentation/devicetree/bindings/ufs/qcom,ufs.yaml | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml b/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml
index 462ead5..d8b807e 100644
--- a/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml
+++ b/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml
@@ -76,6 +76,14 @@ properties:
     $ref: /schemas/types.yaml#/definitions/phandle
     description: phandle to the Inline Crypto Engine node
 
+  qos:
+    minItems: 2
+    maxItems: 2
+    cpumask:
+     description: list of CPUs under CPU group
+    vote:
+     description: vote value for QoS CPU group
+
   reg:
     minItems: 1
     maxItems: 2
@@ -318,5 +326,13 @@ examples:
                             <0 0>,
                             <0 0>;
             qcom,ice = <&ice>;
+            qos0 {
+                cpumask = <0x0f>;
+                vote = <44>;
+            };
+            qos1 {
+                cpumask = <0xf0>;
+                vote = <44>;
+            };
         };
     };
-- 
2.7.4

[PATCH V1 2/4] ufs: ufs-qcom: Add per-cpu PM QoS support for ufs

[Maramaina Naresh]

Qcom SoCs requires registering ufs driver to PM QoS for random io
performance improvement.

PM QoS initialization will parse all QoS CPU group device tree
entries, allocate memory for each individual PM QoS groups and
initiates DEV_PM_QOS_RESUME_LATENCY type of PM QoS request for
new dev_pm_qos_request handle for each CPU. Voting for maximum
latency S32_MAX during driver init.

tiotest benchmark tool io performance results on sm8550 platform:

1. Without PM QoS support
	Type (Speed in)    | Average of 6 iterations
	Random Write(IPOS) | 32201
	Random Read(IPOS)  | 32201

2. With PM QoS support
	Type (Speed in)    | Average of 6 iterations
	Random Write(IPOS) | 40833.5
	Random Read(IPOS)  | 40833.5
(Improvement % with PM QoS = ~20%).

Signed-off-by: Asutosh Das <quic_asutoshd@quicinc.com>
Signed-off-by: Nitin Rawat <quic_nitirawa@quicinc.com>
Signed-off-by: Naveen Kumar Goud Arepalli <quic_narepall@quicinc.com>
Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
---
 drivers/ufs/host/ufs-qcom.c | 152 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/ufs/host/ufs-qcom.h |  29 +++++++++
 2 files changed, 181 insertions(+)

diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
index 2128db0..f3e1e81 100644
--- a/drivers/ufs/host/ufs-qcom.c
+++ b/drivers/ufs/host/ufs-qcom.c
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0-only
 /*
  * Copyright (c) 2013-2016, Linux Foundation. All rights reserved.
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
  */
 
 #include <linux/acpi.h>
@@ -15,6 +16,7 @@
 #include <linux/gpio/consumer.h>
 #include <linux/reset-controller.h>
 #include <linux/devfreq.h>
+#include <linux/cpu.h>
 
 #include <soc/qcom/ice.h>
 
@@ -736,6 +738,53 @@ static void ufs_qcom_device_reset_ctrl(struct ufs_hba *hba, bool asserted)
 	gpiod_set_value_cansleep(host->device_reset, asserted);
 }
 
+/**
+ * ufs_qcom_qos_add_cpu_group - inserts QoS CPU group requests into the QoS list
+ * @qcg: pointer to QoS CPU group
+ * @type: defines the qos request
+ *
+ * Returns 0 for success and non-zero for failure.
+ */
+static int ufs_qcom_qos_add_cpu_group(struct ufs_qcom_qcg *qcg,
+				      enum constraint type)
+{
+	struct dev_pm_qos_request *qos_req = qcg->qos_req;
+	struct device *dev = qcg->host->hba->dev;
+	int cpu;
+	int ret;
+
+	for_each_cpu(cpu, &qcg->mask) {
+		dev_dbg(dev, "cpu: %d | assoc-qos-req: 0x%p\n",
+			cpu, qos_req);
+		ret = dev_pm_qos_add_request(get_cpu_device(cpu),
+					     qos_req,
+					     DEV_PM_QOS_RESUME_LATENCY,
+					     type);
+		if (ret < 0) {
+			dev_err(dev, "Add qos request has failed %d\n",
+				ret);
+			return ret;
+		}
+
+		qos_req++;
+	}
+
+	return 0;
+}
+
+/**
+ * ufs_qcom_qos_remove_cpu_group - removes QoS CPU group requests in QoS list
+ * @qcg: pointer to QoS CPU group
+ */
+static void ufs_qcom_qos_remove_cpu_group(struct ufs_qcom_qcg *qcg)
+{
+	struct dev_pm_qos_request *qos_req = qcg->qos_req;
+	int cpu;
+
+	for_each_cpu(cpu, &qcg->mask)
+		dev_pm_qos_remove_request(qos_req++);
+}
+
 static int ufs_qcom_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op,
 	enum ufs_notify_change_status status)
 {
@@ -1175,6 +1224,107 @@ static int ufs_qcom_icc_init(struct ufs_qcom_host *host)
 }
 
 /**
+ * ufs_qcom_qos_exit - de-allocate QoS instances
+ * @hba: per adapter instance
+ */
+static void ufs_qcom_qos_exit(struct ufs_hba *hba)
+{
+	struct ufs_qcom_host *host = ufshcd_get_variant(hba);
+	struct list_head *head = &host->qos_list_head;
+	struct ufs_qcom_qcg *qcg;
+	struct ufs_qcom_qcg *tmp;
+
+	if (list_empty(head))
+		return;
+
+	list_for_each_entry_safe(qcg, tmp, head, list) {
+		if (qcg->qos_req) {
+			ufs_qcom_qos_remove_cpu_group(qcg);
+			kfree(qcg->qos_req);
+		}
+
+		list_del(&qcg->list);
+		kfree(qcg);
+	}
+}
+
+/**
+ * ufs_qcom_qos_init - initialize QoS instances
+ * @hba: host controller instance
+ *
+ * This function parses all QoS CPU group dt entries, allocates memory for
+ * each individual groups and initiates a qos request for each bit of CPU
+ * mask. As part of init all qos requests are initialized with the maximum
+ * delay aggregated constraint value.
+ */
+static void ufs_qcom_qos_init(struct ufs_hba *hba)
+{
+	struct ufs_qcom_host *host = ufshcd_get_variant(hba);
+	struct device *dev = hba->dev;
+	struct device_node *np = dev->of_node;
+	struct device_node *group_node;
+	struct ufs_qcom_qcg *qcg;
+	cpumask_t mask;
+	char buf[10];
+	u32 cpumask = 0;
+	u32 vote = 0;
+	int ret;
+
+	INIT_LIST_HEAD(&host->qos_list_head);
+
+	for_each_available_child_of_node(np, group_node) {
+		if (!strstr(group_node->name, "qos"))
+			continue;
+
+		of_property_read_u32(group_node, "cpumask", &cpumask);
+		sprintf(buf, "%x", cpumask);
+		cpumask_parse(buf, &mask);
+		if (!cpumask || !cpumask_subset(&mask, cpu_possible_mask)) {
+			dev_err(dev, "Invalid group mask\n");
+			goto out_err;
+		}
+
+		of_property_read_u32(group_node, "vote", &vote);
+		if (!vote) {
+			dev_err(dev, "1 vote is needed, bailing out: %u\n",
+				vote);
+			goto out_err;
+		}
+
+		qcg = kzalloc(sizeof(*qcg), GFP_KERNEL);
+		if (!qcg)
+			goto out_err;
+
+		list_add_tail(&qcg->list, &host->qos_list_head);
+		cpumask_copy(&qcg->mask, &mask);
+		qcg->vote = vote;
+		qcg->host = host;
+		qcg->qos_req = kcalloc(cpumask_weight(&qcg->mask),
+				       sizeof(struct dev_pm_qos_request),
+				       GFP_KERNEL);
+		if (!qcg->qos_req)
+			goto out_err;
+
+		dev_dbg(dev, "qcg: 0x%p | mask-wt: %u | qos_req: 0x%p | vote: %u\n",
+			qcg, cpumask_weight(&qcg->mask),
+			qcg->qos_req, qcg->vote);
+
+		ret = ufs_qcom_qos_add_cpu_group(qcg, S32_MAX);
+		if (ret < 0)
+			goto out_err;
+	}
+
+	if (list_empty(&host->qos_list_head)) {
+		dev_info(dev, "QoS groups undefined\n");
+		return;
+	}
+	return;
+
+out_err:
+	ufs_qcom_qos_exit(hba);
+}
+
+/**
  * ufs_qcom_init - bind phy with controller
  * @hba: host controller instance
  *
@@ -1298,6 +1448,7 @@ static int ufs_qcom_init(struct ufs_hba *hba)
 	 */
 	host->phy_gear = UFS_HS_G2;
 
+	ufs_qcom_qos_init(hba);
 	return 0;
 
 out_variant_clear:
@@ -2036,6 +2187,7 @@ static void ufs_qcom_remove(struct platform_device *pdev)
 	struct ufs_hba *hba =  platform_get_drvdata(pdev);
 
 	pm_runtime_get_sync(&(pdev)->dev);
+	ufs_qcom_qos_exit(hba);
 	ufshcd_remove(hba);
 	platform_msi_domain_free_irqs(hba->dev);
 }
diff --git a/drivers/ufs/host/ufs-qcom.h b/drivers/ufs/host/ufs-qcom.h
index 9950a00..e975a5e 100644
--- a/drivers/ufs/host/ufs-qcom.h
+++ b/drivers/ufs/host/ufs-qcom.h
@@ -1,5 +1,6 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /* Copyright (c) 2013-2015, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
  */
 
 #ifndef UFS_QCOM_H_
@@ -7,6 +8,7 @@
 
 #include <linux/reset-controller.h>
 #include <linux/reset.h>
+#include <linux/pm_qos.h>
 #include <soc/qcom/ice.h>
 #include <ufs/ufshcd.h>
 
@@ -135,6 +137,31 @@ enum {
 #define PA_VS_CORE_CLK_40NS_CYCLES			0x9007
 #define PA_VS_CORE_CLK_40NS_CYCLES_MASK			GENMASK(6, 0)
 
+/**
+ * struct ufs_qcom_qcg - context of QoS cpu group
+ * @qos_req:	pointer to per cpu pm qos request
+ * @host:	pointer to QCOM host controller instance
+ * @list:	helper for linked list
+ * @mask:	parsed cpumask value from device tree
+ * @vote:	parsed vote value from device tree
+ */
+struct ufs_qcom_qcg {
+	struct dev_pm_qos_request *qos_req;
+	struct ufs_qcom_host *host;
+	struct list_head list;
+	cpumask_t mask;
+	u32 vote;
+};
+
+/**
+ * enum constraint - defines QoS constraint type
+ * @QOS_PERF:	QoS performance mode
+ * @QOS_POWER:	QoS power save mode
+ */
+enum constraint {
+	QOS_PERF,
+	QOS_POWER,
+};
 
 /* QCOM UFS host controller core clk frequencies */
 #define UNIPRO_CORE_CLK_FREQ_37_5_MHZ          38
@@ -239,6 +266,8 @@ struct ufs_qcom_host {
 	struct reset_controller_dev rcdev;
 
 	struct gpio_desc *device_reset;
+	/* QoS list head */
+	struct list_head qos_list_head;
 
 	u32 phy_gear;
 
-- 
2.7.4

[PATCH V1 3/4] ufs: ufs-qcom: Add per-cpu PM QoS vote support for ufs

[Maramaina Naresh]

PM QoS per-cpu framework provides performance requirements for each cpu.

The per-cpu PM QoS framework will provide the corresponding interface to
collect the resume_latency request of the specified device and provide
it to the runtime PM. When suspending the device, it will consider this
requirement and decide whether to suspend the device.

Voting will follow below sequence.
1. Vote for maximum latency S32_MAX in driver init.
2. Schedule a vote of PERF when a transfer request is received.
3. Update the vote to S32_MAX during clock gating.

Signed-off-by: Asutosh Das <quic_asutoshd@quicinc.com>
Signed-off-by: Nitin Rawat <quic_nitirawa@quicinc.com>
Signed-off-by: Naveen Kumar Goud Arepalli <quic_narepall@quicinc.com>
Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
---
 drivers/ufs/host/ufs-qcom.c | 188 +++++++++++++++++++++++++++++++++++++++++++-
 drivers/ufs/host/ufs-qcom.h |   8 ++
 2 files changed, 195 insertions(+), 1 deletion(-)

diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
index f3e1e81..2123e8e 100644
--- a/drivers/ufs/host/ufs-qcom.c
+++ b/drivers/ufs/host/ufs-qcom.c
@@ -17,6 +17,7 @@
 #include <linux/reset-controller.h>
 #include <linux/devfreq.h>
 #include <linux/cpu.h>
+#include <linux/blk-mq.h>
 
 #include <soc/qcom/ice.h>
 
@@ -27,6 +28,8 @@
 #include <ufs/ufshci.h>
 #include <ufs/ufs_quirks.h>
 
+#include <scsi/scsi_cmnd.h>
+
 #define MCQ_QCFGPTR_MASK	GENMASK(7, 0)
 #define MCQ_QCFGPTR_UNIT	0x200
 #define MCQ_SQATTR_OFFSET(c) \
@@ -96,6 +99,9 @@ static struct ufs_qcom_host *ufs_qcom_hosts[MAX_UFS_QCOM_HOSTS];
 
 static void ufs_qcom_get_default_testbus_cfg(struct ufs_qcom_host *host);
 static int ufs_qcom_set_core_clk_ctrl(struct ufs_hba *hba, bool is_scale_up);
+static int ufs_qcom_update_qos_constraints(struct ufs_qcom_qcg *qcg,
+					   enum constraint type);
+static int ufs_qcom_qos_unvote_all(struct ufs_qcom_host *host);
 
 static struct ufs_qcom_host *rcdev_to_ufs_host(struct reset_controller_dev *rcd)
 {
@@ -790,6 +796,7 @@ static int ufs_qcom_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op,
 {
 	struct ufs_qcom_host *host = ufshcd_get_variant(hba);
 	struct phy *phy = host->generic_phy;
+	int ret;
 
 	if (status == PRE_CHANGE)
 		return 0;
@@ -810,7 +817,11 @@ static int ufs_qcom_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op,
 		ufs_qcom_disable_lane_clks(host);
 	}
 
-	return ufs_qcom_ice_suspend(host);
+	ret = ufs_qcom_ice_suspend(host);
+	if (ret)
+		return ret;
+
+	return ufs_qcom_qos_unvote_all(host);
 }
 
 static int ufs_qcom_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op)
@@ -840,6 +851,33 @@ static int ufs_qcom_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op)
 	return ufs_qcom_ice_resume(host);
 }
 
+/**
+ * ufs_qcom_qos_unvote_all - unvote QoS for all cpu groups
+ * @host: qcom host controller instance
+ */
+static int ufs_qcom_qos_unvote_all(struct ufs_qcom_host *host)
+{
+	struct list_head *head = &host->qos_list_head;
+	struct ufs_qcom_qcg *qcg;
+	int ret = 0;
+
+	if (list_empty(head))
+		return ret;
+
+	list_for_each_entry(qcg, head, list) {
+		flush_work(&qcg->vwork);
+		if (!qcg->voted)
+			continue;
+
+		ret = ufs_qcom_update_qos_constraints(qcg, QOS_POWER);
+		if (ret)
+			dev_err(host->hba->dev, "Failed to update qos constraints, %d\n",
+				ret);
+	}
+
+	return ret;
+}
+
 static void ufs_qcom_dev_ref_clk_ctrl(struct ufs_qcom_host *host, bool enable)
 {
 	if (host->dev_ref_clk_ctrl_mmio &&
@@ -1157,6 +1195,7 @@ static int ufs_qcom_setup_clocks(struct ufs_hba *hba, bool on,
 		} else {
 			ufs_qcom_icc_set_bw(host, ufs_qcom_bw_table[MODE_MIN][0][0].mem_bw,
 					    ufs_qcom_bw_table[MODE_MIN][0][0].cfg_bw);
+			ufs_qcom_qos_unvote_all(host);
 		}
 		break;
 	}
@@ -1224,6 +1263,144 @@ static int ufs_qcom_icc_init(struct ufs_qcom_host *host)
 }
 
 /**
+ * ufs_qcom_tag_to_cpu - get CPU number for given request tag
+ * @hba: host controller instance
+ * @tag: defines block request id
+ *
+ * Returns 0 or posstive value for success and negative value
+ * for failure.
+ */
+static int ufs_qcom_tag_to_cpu(struct ufs_hba *hba, unsigned int tag)
+{
+	struct ufshcd_lrb *lrbp = &hba->lrb[tag];
+
+	if (lrbp && lrbp->cmd && scsi_cmd_to_rq(lrbp->cmd))
+		return blk_mq_rq_cpu(scsi_cmd_to_rq(lrbp->cmd));
+
+	return -EINVAL;
+}
+
+/**
+ * ufs_qcom_cpu_to_qos_group - returns QoS group address for given CPU number
+ * @host: qcom host controller instance
+ * @cpu: stores CPU number
+ *
+ * Returns ufs_qcom_qcg address for success and NULL for failure.
+ */
+static struct ufs_qcom_qcg *ufs_qcom_cpu_to_qos_group(struct ufs_qcom_host *host,
+						      unsigned int cpu)
+{
+	struct list_head *head = &host->qos_list_head;
+	struct ufs_qcom_qcg *qcg;
+
+	if (cpu > num_possible_cpus())
+		return NULL;
+
+	list_for_each_entry(qcg, head, list) {
+		if (cpumask_test_cpu(cpu, &qcg->mask))
+			return qcg;
+	}
+
+	return NULL;
+}
+
+/**
+ * ufs_qcom_update_qos_constraints - update constraints for QoS group
+ * @qcg: pointer to QoS CPU group
+ * @type: defines the qos request
+ *
+ * Returns 0 for success and non-zero for failure.
+ */
+static int ufs_qcom_update_qos_constraints(struct ufs_qcom_qcg *qcg,
+					   enum constraint type)
+{
+	struct dev_pm_qos_request *qos_req = qcg->qos_req;
+	struct device *dev = qcg->host->hba->dev;
+	unsigned int vote;
+	int cpu;
+	int ret;
+
+	if (type == QOS_POWER)
+		vote = S32_MAX;
+	else
+		vote = qcg->vote;
+
+	if (qcg->curr_vote == vote)
+		return 0;
+
+	for_each_cpu(cpu, &qcg->mask) {
+		dev_dbg(dev, "%s: vote: %d | cpu: %d | qos_req: 0x%p\n",
+			__func__, vote, cpu, qos_req);
+		ret = dev_pm_qos_update_request(qos_req, vote);
+		if (ret < 0)
+			return ret;
+
+		++qos_req;
+	}
+
+	if (type == QOS_POWER)
+		qcg->voted = false;
+	else
+		qcg->voted = true;
+	qcg->curr_vote = vote;
+
+	return 0;
+}
+
+/**
+ * ufs_qcom_setup_xfer_req - setup QoS before transfer request is issued
+ *			and initiates QoS vote process for given tag
+ * @hba: host controller instance
+ * @tag: defines block request id
+ * @is_scsi_cmd: tells scsi cmd or not
+ *
+ * Returns 0 for success and non-zero for failure.
+ */
+static void ufs_qcom_setup_xfer_req(struct ufs_hba *hba, int tag, bool is_scsi_cmd)
+{
+	struct ufs_qcom_host *host = ufshcd_get_variant(hba);
+	struct list_head *head = &host->qos_list_head;
+	struct ufs_qcom_qcg *qcg;
+	int cpu;
+
+	if (list_empty(head))
+		return;
+
+	cpu = ufs_qcom_tag_to_cpu(hba, tag);
+	if (cpu < 0)
+		return;
+
+	qcg = ufs_qcom_cpu_to_qos_group(host, cpu);
+	if (!qcg)
+		return;
+
+	if (qcg->voted) {
+		dev_dbg(hba->dev, "%s: qcg: 0x%p | cpu: %d\n",
+			__func__, qcg, cpu);
+		return;
+	}
+
+	queue_work(host->qos_workq, &qcg->vwork);
+	dev_dbg(hba->dev, "Queued QoS work- cpu: %d\n", cpu);
+}
+
+/**
+ * ufs_qcom_qos_vote_work - starts QoS perf mode vote work
+ * @work: pointer to QoS work
+ */
+static void ufs_qcom_qos_vote_work(struct work_struct *work)
+{
+	struct ufs_qcom_qcg *qcg = container_of(work, struct ufs_qcom_qcg,
+						 vwork);
+	int ret;
+
+	ret = ufs_qcom_update_qos_constraints(qcg, QOS_PERF);
+	if (ret)
+		dev_err(qcg->host->hba->dev, "%s: update qos - failed: %d\n",
+			__func__, ret);
+}
+
+/**
  * ufs_qcom_qos_exit - de-allocate QoS instances
  * @hba: per adapter instance
  */
@@ -1312,12 +1489,20 @@ static void ufs_qcom_qos_init(struct ufs_hba *hba)
 		ret = ufs_qcom_qos_add_cpu_group(qcg, S32_MAX);
 		if (ret < 0)
 			goto out_err;
+
+		INIT_WORK(&qcg->vwork, ufs_qcom_qos_vote_work);
 	}
 
 	if (list_empty(&host->qos_list_head)) {
 		dev_info(dev, "QoS groups undefined\n");
 		return;
 	}
+
+	host->qos_workq = create_singlethread_workqueue("qc_ufs_qos_swq");
+	if (!host->qos_workq) {
+		dev_err(dev, "Failed to create qos workqueue\n");
+		goto out_err;
+	}
 	return;
 
 out_err:
@@ -2148,6 +2333,7 @@ static const struct ufs_hba_variant_ops ufs_hba_qcom_vops = {
 	.dbg_register_dump	= ufs_qcom_dump_dbg_regs,
 	.device_reset		= ufs_qcom_device_reset,
 	.config_scaling_param = ufs_qcom_config_scaling_param,
+	.setup_xfer_req         = ufs_qcom_setup_xfer_req,
 	.program_key		= ufs_qcom_ice_program_key,
 	.reinit_notify		= ufs_qcom_reinit_notify,
 	.mcq_config_resource	= ufs_qcom_mcq_config_resource,
diff --git a/drivers/ufs/host/ufs-qcom.h b/drivers/ufs/host/ufs-qcom.h
index e975a5e..4ad2b3e 100644
--- a/drivers/ufs/host/ufs-qcom.h
+++ b/drivers/ufs/host/ufs-qcom.h
@@ -141,16 +141,22 @@ enum {
  * struct ufs_qcom_qcg - context of QoS cpu group
  * @qos_req:	pointer to per cpu pm qos request
  * @host:	pointer to QCOM host controller instance
+ * @vwork:	qos vote work handle
  * @list:	helper for linked list
  * @mask:	parsed cpumask value from device tree
+ * @curr_vote:	current vote value for group
  * @vote:	parsed vote value from device tree
+ * @voted:	vote status
  */
 struct ufs_qcom_qcg {
 	struct dev_pm_qos_request *qos_req;
 	struct ufs_qcom_host *host;
+	struct work_struct vwork;
 	struct list_head list;
 	cpumask_t mask;
+	u32 curr_vote;
 	u32 vote;
+	bool voted;
 };
 
 /**
@@ -268,6 +274,8 @@ struct ufs_qcom_host {
 	struct gpio_desc *device_reset;
 	/* QoS list head */
 	struct list_head qos_list_head;
+	/* QoS workqueue */
+	struct workqueue_struct *qos_workq;
 
 	u32 phy_gear;
 
-- 
2.7.4

[PATCH V1 4/4] arm64: dts: qcom: sm8550: Add per-cpu PM QoS support for ufs

[Maramaina Naresh]

Add per-cpu PM QoS support for ufs.

Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
---
 arch/arm64/boot/dts/qcom/sm8550.dtsi | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/sm8550.dtsi b/arch/arm64/boot/dts/qcom/sm8550.dtsi
index 7b9ddde..84909b0 100644
--- a/arch/arm64/boot/dts/qcom/sm8550.dtsi
+++ b/arch/arm64/boot/dts/qcom/sm8550.dtsi
@@ -1956,6 +1956,15 @@
 			qcom,ice = <&ice>;
 
 			status = "disabled";
+
+			qos0 {
+				cpumask = <0x0f>;
+				vote = <44>;
+			};
+			qos1 {
+				cpumask = <0xf0>;
+				vote = <44>;
+			};
 		};
 
 		ice: crypto@1d88000 {
-- 
2.7.4

Re: [PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property

[Rob Herring]

On Tue, 10 Oct 2023 21:19:04 +0530, Maramaina Naresh wrote:
> Add bindings for per-cpu QoS for QCOM UFS. This improves random io
> performance by 20% for QCOM UFS.
> 
> Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
> ---
>  Documentation/devicetree/bindings/ufs/qcom,ufs.yaml | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 

My bot found errors running 'make DT_CHECKER_FLAGS=-m dt_binding_check'
on your patch (DT_CHECKER_FLAGS is new in v5.13):

yamllint warnings/errors:
./Documentation/devicetree/bindings/ufs/qcom,ufs.yaml:83:6: [warning] wrong indentation: expected 6 but found 5 (indentation)
./Documentation/devicetree/bindings/ufs/qcom,ufs.yaml:85:6: [warning] wrong indentation: expected 6 but found 5 (indentation)

dtschema/dtc warnings/errors:
/builds/robherring/dt-review-ci/linux/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml: properties:qos: 'anyOf' conditional failed, one must be fixed:
	'cpumask' is not one of ['$ref', 'additionalItems', 'additionalProperties', 'allOf', 'anyOf', 'const', 'contains', 'default', 'dependencies', 'dependentRequired', 'dependentSchemas', 'deprecated', 'description', 'else', 'enum', 'exclusiveMaximum', 'exclusiveMinimum', 'items', 'if', 'minItems', 'minimum', 'maxItems', 'maximum', 'multipleOf', 'not', 'oneOf', 'pattern', 'patternProperties', 'properties', 'required', 'then', 'typeSize', 'unevaluatedProperties', 'uniqueItems']
	'type' was expected
	from schema $id: http://devicetree.org/meta-schemas/keywords.yaml#
/builds/robherring/dt-review-ci/linux/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml: properties:qos: 'anyOf' conditional failed, one must be fixed:
	'vote' is not one of ['$ref', 'additionalItems', 'additionalProperties', 'allOf', 'anyOf', 'const', 'contains', 'default', 'dependencies', 'dependentRequired', 'dependentSchemas', 'deprecated', 'description', 'else', 'enum', 'exclusiveMaximum', 'exclusiveMinimum', 'items', 'if', 'minItems', 'minimum', 'maxItems', 'maximum', 'multipleOf', 'not', 'oneOf', 'pattern', 'patternProperties', 'properties', 'required', 'then', 'typeSize', 'unevaluatedProperties', 'uniqueItems']
	'type' was expected
	from schema $id: http://devicetree.org/meta-schemas/keywords.yaml#
/builds/robherring/dt-review-ci/linux/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml: qos: missing type definition
/builds/robherring/dt-review-ci/linux/Documentation/devicetree/bindings/ufs/qcom,ufs.example.dtb: ufs@1d84000: Unevaluated properties are not allowed ('qos0', 'qos1' were unexpected)
	from schema $id: http://devicetree.org/schemas/ufs/qcom,ufs.yaml#

doc reference errors (make refcheckdocs):

See https://patchwork.ozlabs.org/project/devicetree-bindings/patch/1696952947-18062-2-git-send-email-quic_mnaresh@quicinc.com

The base for the series is generally the latest rc1. A different dependency
should be noted in *this* patch.

If you already ran 'make dt_binding_check' and didn't see the above
error(s), then make sure 'yamllint' is installed and dt-schema is up to
date:

pip3 install dtschema --upgrade

Please check and re-submit after running the above command yourself. Note
that DT_SCHEMA_FILES can be set to your schema file to speed up checking
your schema. However, it must be unset to test all examples with your schema.

Re: [PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property

[Konrad Dybcio]

On 10/10/23 17:49, Maramaina Naresh wrote:
> Add bindings for per-cpu QoS for QCOM UFS. This improves random io
> performance by 20% for QCOM UFS.
The bindings addition improves performance? Cool :P

Please explain your changes and provide the "why".

[...]

> @@ -318,5 +326,13 @@ examples:
>                               <0 0>,
>                               <0 0>;
>               qcom,ice = <&ice>;
> +            qos0 {
Looks like the indentation is off, also missing newline before subnodes.

> +                cpumask = <0x0f>;This should be a CPU phandle array instead. Besides, can we not 
determine this dynamically?

> +                vote = <44>;
> +            };
> +            qos1 {
Missing newline between subnodes.

Konrad

Re: [PATCH V1 3/4] ufs: ufs-qcom: Add per-cpu PM QoS vote support for ufs

[Konrad Dybcio]

On 10/10/23 17:49, Maramaina Naresh wrote:
> PM QoS per-cpu framework provides performance requirements for each cpu.
> 
> The per-cpu PM QoS framework will provide the corresponding interface to
> collect the resume_latency request of the specified device and provide
> it to the runtime PM. When suspending the device, it will consider this
> requirement and decide whether to suspend the device.
> 
> Voting will follow below sequence.
> 1. Vote for maximum latency S32_MAX in driver init.
> 2. Schedule a vote of PERF when a transfer request is received.
> 3. Update the vote to S32_MAX during clock gating.
> 
> Signed-off-by: Asutosh Das <quic_asutoshd@quicinc.com>
> Signed-off-by: Nitin Rawat <quic_nitirawa@quicinc.com>
> Signed-off-by: Naveen Kumar Goud Arepalli <quic_narepall@quicinc.com>
> Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
> ---
Is this the same patch as 2/4?

Konrad

Re: [PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property

[kernel test robot]

Hi Maramaina,

kernel test robot noticed the following build warnings:

[auto build test WARNING on mkp-scsi/for-next]
[also build test WARNING on jejb-scsi/for-next robh/for-next linus/master v6.6-rc5 next-20231010]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Maramaina-Naresh/dt-bindings-ufs-qcom-Add-qos-property/20231010-235602
base:   https://git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi.git for-next
patch link:    https://lore.kernel.org/r/1696952947-18062-2-git-send-email-quic_mnaresh%40quicinc.com
patch subject: [PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property
compiler: loongarch64-linux-gcc (GCC) 13.2.0
reproduce: (https://download.01.org/0day-ci/archive/20231011/202310111509.fFHOVdaI-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202310111509.fFHOVdaI-lkp@intel.com/

dtcheck warnings: (new ones prefixed by >>)
>> Documentation/devicetree/bindings/ufs/qcom,ufs.yaml:83:6: [warning] wrong indentation: expected 6 but found 5 (indentation)
   Documentation/devicetree/bindings/ufs/qcom,ufs.yaml:85:6: [warning] wrong indentation: expected 6 but found 5 (indentation)
--
>> Documentation/devicetree/bindings/ufs/qcom,ufs.yaml: properties:qos: 'anyOf' conditional failed, one must be fixed:
   	'cpumask' is not one of ['$ref', 'additionalItems', 'additionalProperties', 'allOf', 'anyOf', 'const', 'contains', 'default', 'dependencies', 'dependentRequired', 'dependentSchemas', 'deprecated', 'description', 'else', 'enum', 'exclusiveMaximum', 'exclusiveMinimum', 'items', 'if', 'minItems', 'minimum', 'maxItems', 'maximum', 'multipleOf', 'not', 'oneOf', 'pattern', 'patternProperties', 'properties', 'required', 'then', 'typeSize', 'unevaluatedProperties', 'uniqueItems']
   	'type' was expected
   	from schema $id: http://devicetree.org/meta-schemas/keywords.yaml#
>> Documentation/devicetree/bindings/ufs/qcom,ufs.yaml: properties:qos: 'anyOf' conditional failed, one must be fixed:
   	'vote' is not one of ['$ref', 'additionalItems', 'additionalProperties', 'allOf', 'anyOf', 'const', 'contains', 'default', 'dependencies', 'dependentRequired', 'dependentSchemas', 'deprecated', 'description', 'else', 'enum', 'exclusiveMaximum', 'exclusiveMinimum', 'items', 'if', 'minItems', 'minimum', 'maxItems', 'maximum', 'multipleOf', 'not', 'oneOf', 'pattern', 'patternProperties', 'properties', 'required', 'then', 'typeSize', 'unevaluatedProperties', 'uniqueItems']
   	'type' was expected
   	from schema $id: http://devicetree.org/meta-schemas/keywords.yaml#

vim +83 Documentation/devicetree/bindings/ufs/qcom,ufs.yaml

     8	
     9	maintainers:
    10	  - Bjorn Andersson <bjorn.andersson@linaro.org>
    11	  - Andy Gross <agross@kernel.org>
    12	
    13	# Select only our matches, not all jedec,ufs-2.0
    14	select:
    15	  properties:
    16	    compatible:
    17	      contains:
    18	        const: qcom,ufshc
    19	  required:
    20	    - compatible
    21	
    22	properties:
    23	  compatible:
    24	    items:
    25	      - enum:
    26	          - qcom,msm8994-ufshc
    27	          - qcom,msm8996-ufshc
    28	          - qcom,msm8998-ufshc
    29	          - qcom,sa8775p-ufshc
    30	          - qcom,sc8280xp-ufshc
    31	          - qcom,sdm845-ufshc
    32	          - qcom,sm6115-ufshc
    33	          - qcom,sm6350-ufshc
    34	          - qcom,sm8150-ufshc
    35	          - qcom,sm8250-ufshc
    36	          - qcom,sm8350-ufshc
    37	          - qcom,sm8450-ufshc
    38	          - qcom,sm8550-ufshc
    39	      - const: qcom,ufshc
    40	      - const: jedec,ufs-2.0
    41	
    42	  clocks:
    43	    minItems: 8
    44	    maxItems: 11
    45	
    46	  clock-names:
    47	    minItems: 8
    48	    maxItems: 11
    49	
    50	  dma-coherent: true
    51	
    52	  interconnects:
    53	    minItems: 2
    54	    maxItems: 2
    55	
    56	  interconnect-names:
    57	    items:
    58	      - const: ufs-ddr
    59	      - const: cpu-ufs
    60	
    61	  iommus:
    62	    minItems: 1
    63	    maxItems: 2
    64	
    65	  phys:
    66	    maxItems: 1
    67	
    68	  phy-names:
    69	    items:
    70	      - const: ufsphy
    71	
    72	  power-domains:
    73	    maxItems: 1
    74	
    75	  qcom,ice:
    76	    $ref: /schemas/types.yaml#/definitions/phandle
    77	    description: phandle to the Inline Crypto Engine node
    78	
    79	  qos:
    80	    minItems: 2
    81	    maxItems: 2
    82	    cpumask:
  > 83	     description: list of CPUs under CPU group
    84	    vote:
    85	     description: vote value for QoS CPU group
    86	
    87	  reg:
    88	    minItems: 1
    89	    maxItems: 2
    90	
    91	  reg-names:
    92	    items:
    93	      - const: std
    94	      - const: ice
    95	
    96	  required-opps:
    97	    maxItems: 1
    98	
    99	  resets:
   100	    maxItems: 1
   101	
   102	  '#reset-cells':
   103	    const: 1
   104	
   105	  reset-names:
   106	    items:
   107	      - const: rst
   108	
   109	  reset-gpios:
   110	    maxItems: 1
   111	    description:
   112	      GPIO connected to the RESET pin of the UFS memory device.
   113	
   114	required:
   115	  - compatible
   116	  - reg
   117	
   118	allOf:
   119	  - $ref: ufs-common.yaml
   120	
   121	  - if:
   122	      properties:
   123	        compatible:
   124	          contains:
   125	            enum:
   126	              - qcom,msm8998-ufshc
   127	              - qcom,sa8775p-ufshc
   128	              - qcom,sc8280xp-ufshc
   129	              - qcom,sm8250-ufshc
   130	              - qcom,sm8350-ufshc
   131	              - qcom,sm8450-ufshc
   132	              - qcom,sm8550-ufshc
   133	    then:
   134	      properties:
   135	        clocks:
   136	          minItems: 8
   137	          maxItems: 8
   138	        clock-names:
   139	          items:
   140	            - const: core_clk
   141	            - const: bus_aggr_clk
   142	            - const: iface_clk
   143	            - const: core_clk_unipro
   144	            - const: ref_clk
   145	            - const: tx_lane0_sync_clk
   146	            - const: rx_lane0_sync_clk
   147	            - const: rx_lane1_sync_clk
   148	        reg:
   149	          minItems: 1
   150	          maxItems: 1
   151	        reg-names:
   152	          maxItems: 1
   153	
   154	  - if:
   155	      properties:
   156	        compatible:
   157	          contains:
   158	            enum:
   159	              - qcom,sdm845-ufshc
   160	              - qcom,sm6350-ufshc
   161	              - qcom,sm8150-ufshc
   162	    then:
   163	      properties:
   164	        clocks:
   165	          minItems: 9
   166	          maxItems: 9
   167	        clock-names:
   168	          items:
   169	            - const: core_clk
   170	            - const: bus_aggr_clk
   171	            - const: iface_clk
   172	            - const: core_clk_unipro
   173	            - const: ref_clk
   174	            - const: tx_lane0_sync_clk
   175	            - const: rx_lane0_sync_clk
   176	            - const: rx_lane1_sync_clk
   177	            - const: ice_core_clk
   178	        reg:
   179	          minItems: 2
   180	          maxItems: 2
   181	        reg-names:
   182	          minItems: 2
   183	      required:
   184	        - reg-names
   185	
   186	  - if:
   187	      properties:
   188	        compatible:
   189	          contains:
   190	            enum:
   191	              - qcom,msm8996-ufshc
   192	    then:
   193	      properties:
   194	        clocks:
   195	          minItems: 11
   196	          maxItems: 11
   197	        clock-names:
   198	          items:
   199	            - const: core_clk_src
   200	            - const: core_clk
   201	            - const: bus_clk
   202	            - const: bus_aggr_clk
   203	            - const: iface_clk
   204	            - const: core_clk_unipro_src
   205	            - const: core_clk_unipro
   206	            - const: core_clk_ice
   207	            - const: ref_clk
   208	            - const: tx_lane0_sync_clk
   209	            - const: rx_lane0_sync_clk
   210	        reg:
   211	          minItems: 1
   212	          maxItems: 1
   213	        reg-names:
   214	          maxItems: 1
   215	
   216	  - if:
   217	      properties:
   218	        compatible:
   219	          contains:
   220	            enum:
   221	              - qcom,sm6115-ufshc
   222	    then:
   223	      properties:
   224	        clocks:
   225	          minItems: 8
   226	          maxItems: 8
   227	        clock-names:
   228	          items:
   229	            - const: core_clk
   230	            - const: bus_aggr_clk
   231	            - const: iface_clk
   232	            - const: core_clk_unipro
   233	            - const: ref_clk
   234	            - const: tx_lane0_sync_clk
   235	            - const: rx_lane0_sync_clk
   236	            - const: ice_core_clk
   237	        reg:
   238	          minItems: 2
   239	          maxItems: 2
   240	        reg-names:
   241	          minItems: 2
   242	      required:
   243	        - reg-names
   244	
   245	    # TODO: define clock bindings for qcom,msm8994-ufshc
   246	
   247	  - if:
   248	      required:
   249	        - qcom,ice
   250	    then:
   251	      properties:
   252	        reg:
   253	          maxItems: 1
   254	        clocks:
   255	          minItems: 8
   256	          maxItems: 8
   257	    else:
   258	      properties:
   259	        reg:
   260	          minItems: 1
   261	          maxItems: 2
   262	        clocks:
   263	          minItems: 8
   264	          maxItems: 11
   265	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH V1 0/4] Add per-cpu PM QoS support for QCOM UFS

[Bart Van Assche]

On 10/10/23 08:49, Maramaina Naresh wrote:
> Add per-cpu PM QoS support for ufs. This improves random io performance
> by 20% for ufs.

What in this patch series (other than the DT-bindings) is specific to
the Qualcomm driver? If the answer is not much: please move this
functionality into the UFS driver core.

Thanks,

Bart.

Re: [PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property

[Krzysztof Kozlowski]

On 10/10/2023 17:49, Maramaina Naresh wrote:
> Add bindings for per-cpu QoS for QCOM UFS. This improves random io
> performance by 20% for QCOM UFS.
> 
> Signed-off-by: Maramaina Naresh <quic_mnaresh@quicinc.com>
> ---
>  Documentation/devicetree/bindings/ufs/qcom,ufs.yaml | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml b/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml
> index 462ead5..d8b807e 100644
> --- a/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml
> +++ b/Documentation/devicetree/bindings/ufs/qcom,ufs.yaml
> @@ -76,6 +76,14 @@ properties:
>      $ref: /schemas/types.yaml#/definitions/phandle
>      description: phandle to the Inline Crypto Engine node
>  
> +  qos:
> +    minItems: 2
> +    maxItems: 2

This was absolutely never tested.

Best regards,
Krzysztof

Re: [PATCH V1 0/4] Add per-cpu PM QoS support for QCOM UFS

[Maramaina Naresh]

Hi Bart,

Thank you for reviewing the patch.  This is not specific to the Qualcomm 
driver.

We will move this feature to core UFS.

Thanks,

Naresh.

On 10/12/2023 2:26 AM, Bart Van Assche wrote:
> On 10/10/23 08:49, Maramaina Naresh wrote:
>> Add per-cpu PM QoS support for ufs. This improves random io performance
>> by 20% for ufs.
>
> What in this patch series (other than the DT-bindings) is specific to
> the Qualcomm driver? If the answer is not much: please move this
> functionality into the UFS driver core.
>
> Thanks,
>
> Bart.
>

Re: [PATCH V1 1/4] dt-bindings: ufs: qcom: Add qos property

[Konrad Dybcio]

On 10/20/23 11:04, Maramaina Naresh wrote:
> Hi All,
> 
> Thanks for reviewing the patch.
> 
> We are moving this feature support to core UFS driver and planning to control through UFS capability flag [ As per upstream comments ].
> 
> We will take care of comments and push new changes in next patch series.
Sounds good.

Please also fix your email client to avoid HTML. The lists will
only accept plaintext messages.

Konrad