Re: [PATCH V2 1/3] ufs: core: Add CPU latency QoS support for ufs driver
From: Naresh Maramaina
Date: Mon Dec 11 2023 - 04:57:13 EST
On 12/6/2023 8:56 PM, Manivannan Sadhasivam wrote:
On Mon, Dec 04, 2023 at 08:00:59PM +0530, Maramaina Naresh wrote:
Register ufs driver to CPU latency PM QoS framework can improves
ufs device random io performance.
PM QoS initialization will insert new QoS request into the CPU
latency QoS list with the maximum latency PM_QOS_DEFAULT_VALUE
value.
UFS driver will vote for performance mode on scale up and power
save mode for scale down.
If clock scaling feature is not enabled then voting will be based
on clock on or off condition.
tiotest benchmark tool io performance results on sm8550 platform:
1. Without PM QoS support
Type (Speed in) | Average of 18 iterations
Random Write(IPOS) | 41065.13
Random Read(IPOS) | 37101.3
2. With PM QoS support
Type (Speed in) | Average of 18 iterations
Random Write(IPOS) | 46784.9
Random Read(IPOS) | 42943.4
(Improvement % with PM QoS = ~15%).
Co-developed-by: Nitin Rawat <quic_nitirawa@xxxxxxxxxxx>
Signed-off-by: Nitin Rawat <quic_nitirawa@xxxxxxxxxxx>
Signed-off-by: Naveen Kumar Goud Arepalli <quic_narepall@xxxxxxxxxxx>
Signed-off-by: Maramaina Naresh <quic_mnaresh@xxxxxxxxxxx>
---
drivers/ufs/core/ufshcd-priv.h | 8 +++++
drivers/ufs/core/ufshcd.c | 62 ++++++++++++++++++++++++++++++++++
include/ufs/ufshcd.h | 16 +++++++++
3 files changed, 86 insertions(+)
diff --git a/drivers/ufs/core/ufshcd-priv.h b/drivers/ufs/core/ufshcd-priv.h
index f42d99ce5bf1..536805f6c4e1 100644
--- a/drivers/ufs/core/ufshcd-priv.h
+++ b/drivers/ufs/core/ufshcd-priv.h
@@ -241,6 +241,14 @@ static inline void ufshcd_vops_config_scaling_param(struct ufs_hba *hba,
hba->vops->config_scaling_param(hba, p, data);
}
+static inline u32 ufshcd_vops_config_qos_vote(struct ufs_hba *hba)
+{
+ if (hba->vops && hba->vops->config_qos_vote)
+ return hba->vops->config_qos_vote(hba);
Please remove this callback as Bart noted.
Sure Mani, will takecare of this comment.
+
+ return UFSHCD_QOS_DEFAULT_VOTE;
+}
+
static inline void ufshcd_vops_reinit_notify(struct ufs_hba *hba)
{
if (hba->vops && hba->vops->reinit_notify)
diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index ae9936fc6ffb..13370febd2b5 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -1001,6 +1001,20 @@ static bool ufshcd_is_unipro_pa_params_tuning_req(struct ufs_hba *hba)
return ufshcd_get_local_unipro_ver(hba) < UFS_UNIPRO_VER_1_6;
}
+/**
+ * ufshcd_pm_qos_perf - vote for PM QoS performance or power save mode
ufshcd_pm_qos_update() - Update PM QoS request
Sure Mani, will takecare of this comment.
+ * @hba: per adapter instance
+ * @on: If True, vote for perf PM QoS mode otherwise power save mode
+ */
+static void ufshcd_pm_qos_perf(struct ufs_hba *hba, bool on)
+{
+ if (!hba->pm_qos_init)
+ return;
+
+ cpu_latency_qos_update_request(&hba->pm_qos_req, on ? hba->qos_vote
+ : PM_QOS_DEFAULT_VALUE);
+}
+
/**
* ufshcd_set_clk_freq - set UFS controller clock frequencies
* @hba: per adapter instance
@@ -1153,6 +1167,10 @@ static int ufshcd_scale_clks(struct ufs_hba *hba, unsigned long freq,
trace_ufshcd_profile_clk_scaling(dev_name(hba->dev),
(scale_up ? "up" : "down"),
ktime_to_us(ktime_sub(ktime_get(), start)), ret);
+
+ if (!ret)
+ ufshcd_pm_qos_perf(hba, scale_up);
Can't you just move this before trace_ufshcd_profile_clk_scaling()? This also
avoids checking for !ret.
In this case, we need to use goto out; inside if condition of
ufshcd_vops_clk_scale_notify.
we can enable ufshcd_pm_qos_perf only when ufshcd_vops_clk_scale_notify
is successful.
Will add goto out; in next patch set.
+
return ret;
}
@@ -9204,6 +9222,8 @@ static int ufshcd_setup_clocks(struct ufs_hba *hba, bool on)
if (ret)
return ret;
+ if (!ufshcd_is_clkscaling_supported(hba))
+ ufshcd_pm_qos_perf(hba, on);
out:
if (ret) {
list_for_each_entry(clki, head, list) {
@@ -9296,6 +9316,45 @@ static int ufshcd_init_clocks(struct ufs_hba *hba)
return ret;
}
+/**
+ * ufshcd_pm_qos_init - initialize PM QoS instance
"Initialize PM QoS request"
Sure Mani, will takecare of this comment.
+ * @hba: per adapter instance
+ */
+static void ufshcd_pm_qos_init(struct ufs_hba *hba)
+{
+ if (!(hba->caps & UFSHCD_CAP_PM_QOS))
+ return;
+
+ /*
+ * called to configure PM QoS vote value for UFS host,
+ * expecting qos vote return value from caller else
+ * default vote value will be return.
+ */
+ hba->qos_vote = ufshcd_vops_config_qos_vote(hba);
No need of this variable too if you get rid of the callback.
+ cpu_latency_qos_add_request(&hba->pm_qos_req,
+ PM_QOS_DEFAULT_VALUE);
+
+ if (cpu_latency_qos_request_active(&hba->pm_qos_req))
+ hba->pm_qos_init = true;
Why do you need this flag?
this flag ensure UFS qos request got added into the list.
+
+ dev_dbg(hba->dev, "%s: QoS %s, qos_vote: %u\n", __func__,
+ hba->pm_qos_init ? "initialized" : "uninitialized",
+ hba->qos_vote);
+}
+
+/**
+ * ufshcd_pm_qos_exit - remove instance from PM QoS
+ * @hba: per adapter instance
+ */
+static void ufshcd_pm_qos_exit(struct ufs_hba *hba)
+{
+ if (!hba->pm_qos_init)
+ return;
+
+ cpu_latency_qos_remove_request(&hba->pm_qos_req);
+ hba->pm_qos_init = false;
+}
+
[...]
/**
* struct ufs_hba - per adapter private structure
* @mmio_base: UFSHCI base register address
@@ -912,6 +923,8 @@ enum ufshcd_mcq_opr {
* @mcq_base: Multi circular queue registers base address
* @uhq: array of supported hardware queues
* @dev_cmd_queue: Queue for issuing device management commands
+ * @pm_qos_req: PM QoS request handle
+ * @pm_qos_init: flag to check if pm qos init completed
*/
struct ufs_hba {
void __iomem *mmio_base;
@@ -1076,6 +1089,9 @@ struct ufs_hba {
struct ufs_hw_queue *uhq;
struct ufs_hw_queue *dev_cmd_queue;
struct ufshcd_mcq_opr_info_t mcq_opr[OPR_MAX];
+ struct pm_qos_request pm_qos_req;
+ bool pm_qos_init;
+ u32 qos_vote;
Order doesn't match Kdoc.
qos_vote variable will be remove as per latest comment.
- Mani
Thanks,
Naresh.