Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates
 216.228.117.160 as permitted sender) receiver=protection.outlook.com;
 client-ip=216.228.117.160; helo=mail.nvidia.com; pr=C
From:   Sumit Gupta <sumitg@nvidia.com>
To:     <rafael@kernel.org>, <viresh.kumar@linaro.org>,
        <linux-pm@vger.kernel.org>, <linux-tegra@vger.kernel.org>,
        <linux-kernel@vger.kernel.org>
CC:     <treding@nvidia.com>, <jonathanh@nvidia.com>, <bbasu@nvidia.com>,
        <sumitg@nvidia.com>, <amiettinen@nvidia.com>
Subject: [Patch v3 2/2] cpufreq: tegra194: use refclk delta based loop instead of udelay
Date:   Wed, 4 Oct 2023 19:35:37 +0530
Message-ID: <20231004140537.1954-3-sumitg@nvidia.com>
In-Reply-To: <20231004140537.1954-1-sumitg@nvidia.com>
References: <20231004140537.1954-1-sumitg@nvidia.com>
MIME-Version: 1.0
Content-Type: text/plain
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 04 Oct 2023 14:06:07.8352
 (UTC)
X-MS-Exchange-CrossTenant-Network-Message-Id: c91a4ded-7031-4a20-956b-08dbc4e30e4a
X-MS-Exchange-CrossTenant-Id: 43083d15-7273-40c1-b7db-39efd9ccc17a
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=43083d15-7273-40c1-b7db-39efd9ccc17a;Ip=[216.228.117.160];Helo=[mail.nvidia.com]
X-MS-Exchange-CrossTenant-AuthSource: CY4PEPF0000EE37.namprd05.prod.outlook.com
X-MS-Exchange-CrossTenant-AuthAs: Anonymous
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: DM4PR12MB6183
X-Spam-Status: No, score=-1.6 required=5.0 tests=BAYES_00,DKIMWL_WL_HIGH,
        DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,FORGED_SPF_HELO,
        RCVD_IN_DNSWL_BLOCKED,RCVD_IN_MSPIKE_H2,SPF_HELO_PASS,SPF_NONE,
        URIBL_BLOCKED autolearn=no autolearn_force=no version=3.4.6
X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on
        lindbergh.monkeyblade.net
Precedence: bulk
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
X-Greylist: Sender passed SPF test, not delayed by milter-greylist-4.6.4 (snail.vger.email [0.0.0.0]); Wed, 04 Oct 2023 07:06:16 -0700 (PDT)

Use reference clock count based loop instead of "udelay()" for
sampling of counters to improve the accuracy of re-generated CPU
frequency. "udelay()" internally calls "WFE" which stops the
counters and results in bigger delta between the last set freq
and the re-generated value from counters. The counter sampling
window used in loop is the minimum number of reference clock
cycles which is known to give a stable value of CPU frequency.
The change also helps to reduce the sampling window from "500us"
to "<50us".

Suggested-by: Antti Miettinen <amiettinen@nvidia.com>
Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
---
 drivers/cpufreq/tegra194-cpufreq.c | 72 +++++++++++++++++++++++-------
 1 file changed, 55 insertions(+), 17 deletions(-)

diff --git a/drivers/cpufreq/tegra194-cpufreq.c b/drivers/cpufreq/tegra194-cpufreq.c
index f6a8e6cf6d94..9dae6195e0e7 100644
--- a/drivers/cpufreq/tegra194-cpufreq.c
+++ b/drivers/cpufreq/tegra194-cpufreq.c
@@ -5,7 +5,6 @@
 
 #include <linux/cpu.h>
 #include <linux/cpufreq.h>
-#include <linux/delay.h>
 #include <linux/dma-mapping.h>
 #include <linux/module.h>
 #include <linux/of.h>
@@ -21,10 +20,11 @@
 
 #define KHZ                     1000
 #define REF_CLK_MHZ             408 /* 408 MHz */
-#define US_DELAY                500
 #define CPUFREQ_TBL_STEP_HZ     (50 * KHZ * KHZ)
 #define MAX_CNT                 ~0U
 
+#define MAX_DELTA_KHZ          115200
+
 #define NDIV_MASK              0x1FF
 
 #define CORE_OFFSET(cpu)			(cpu * 8)
@@ -68,6 +68,7 @@ struct tegra_cpufreq_soc {
 	int maxcpus_per_cluster;
 	unsigned int num_clusters;
 	phys_addr_t actmon_cntr_base;
+	u32 refclk_delta_min;
 };
 
 struct tegra194_cpufreq_data {
@@ -149,6 +150,8 @@ static void tegra234_read_counters(struct tegra_cpu_ctr *c)
 {
 	struct tegra194_cpufreq_data *data = cpufreq_get_driver_data();
 	void __iomem *actmon_reg;
+	u32 delta_refcnt;
+	int cnt = 0;
 	u64 val;
 
 	actmon_reg = CORE_ACTMON_CNTR_REG(data, data->cpu_data[c->cpu].clusterid,
@@ -157,10 +160,25 @@ static void tegra234_read_counters(struct tegra_cpu_ctr *c)
 	val = readq(actmon_reg);
 	c->last_refclk_cnt = upper_32_bits(val);
 	c->last_coreclk_cnt = lower_32_bits(val);
-	udelay(US_DELAY);
-	val = readq(actmon_reg);
-	c->refclk_cnt = upper_32_bits(val);
-	c->coreclk_cnt = lower_32_bits(val);
+
+	/*
+	 * The sampling window is based on the minimum number of reference
+	 * clock cycles which is known to give a stable value of CPU frequency.
+	 */
+	do {
+		val = readq(actmon_reg);
+		c->refclk_cnt = upper_32_bits(val);
+		c->coreclk_cnt = lower_32_bits(val);
+		if (c->refclk_cnt < c->last_refclk_cnt)
+			delta_refcnt = c->refclk_cnt + (MAX_CNT - c->last_refclk_cnt);
+		else
+			delta_refcnt = c->refclk_cnt - c->last_refclk_cnt;
+		if (++cnt >= 0xFFFF) {
+			pr_warn("cpufreq: problem with refclk on cpu:%d, delta_refcnt:%u, cnt:%d\n",
+				c->cpu, delta_refcnt, cnt);
+			break;
+		}
+	} while (delta_refcnt < data->soc->refclk_delta_min);
 }
 
 static struct tegra_cpufreq_ops tegra234_cpufreq_ops = {
@@ -175,6 +193,7 @@ static const struct tegra_cpufreq_soc tegra234_cpufreq_soc = {
 	.actmon_cntr_base = 0x9000,
 	.maxcpus_per_cluster = 4,
 	.num_clusters = 3,
+	.refclk_delta_min = 16000,
 };
 
 static const struct tegra_cpufreq_soc tegra239_cpufreq_soc = {
@@ -182,6 +201,7 @@ static const struct tegra_cpufreq_soc tegra239_cpufreq_soc = {
 	.actmon_cntr_base = 0x4000,
 	.maxcpus_per_cluster = 8,
 	.num_clusters = 1,
+	.refclk_delta_min = 16000,
 };
 
 static void tegra194_get_cpu_cluster_id(u32 cpu, u32 *cpuid, u32 *clusterid)
@@ -222,15 +242,33 @@ static inline u32 map_ndiv_to_freq(struct mrq_cpu_ndiv_limits_response
 
 static void tegra194_read_counters(struct tegra_cpu_ctr *c)
 {
+	struct tegra194_cpufreq_data *data = cpufreq_get_driver_data();
+	u32 delta_refcnt;
+	int cnt = 0;
 	u64 val;
 
 	val = read_freq_feedback();
 	c->last_refclk_cnt = lower_32_bits(val);
 	c->last_coreclk_cnt = upper_32_bits(val);
-	udelay(US_DELAY);
-	val = read_freq_feedback();
-	c->refclk_cnt = lower_32_bits(val);
-	c->coreclk_cnt = upper_32_bits(val);
+
+	/*
+	 * The sampling window is based on the minimum number of reference
+	 * clock cycles which is known to give a stable value of CPU frequency.
+	 */
+	do {
+		val = read_freq_feedback();
+		c->refclk_cnt = lower_32_bits(val);
+		c->coreclk_cnt = upper_32_bits(val);
+		if (c->refclk_cnt < c->last_refclk_cnt)
+			delta_refcnt = c->refclk_cnt + (MAX_CNT - c->last_refclk_cnt);
+		else
+			delta_refcnt = c->refclk_cnt - c->last_refclk_cnt;
+		if (++cnt >= 0xFFFF) {
+			pr_warn("cpufreq: problem with refclk on cpu:%d, delta_refcnt:%u, cnt:%d\n",
+				c->cpu, delta_refcnt, cnt);
+			break;
+		}
+	} while (delta_refcnt < data->soc->refclk_delta_min);
 }
 
 static void tegra_read_counters(struct work_struct *work)
@@ -288,9 +326,8 @@ static unsigned int tegra194_calculate_speed(u32 cpu)
 	u32 rate_mhz;
 
 	/*
-	 * udelay() is required to reconstruct cpu frequency over an
-	 * observation window. Using workqueue to call udelay() with
-	 * interrupts enabled.
+	 * Reconstruct cpu frequency over an observation/sampling window.
+	 * Using workqueue to keep interrupts enabled during the interval.
 	 */
 	read_counters_work.c.cpu = cpu;
 	INIT_WORK_ONSTACK(&read_counters_work.work, tegra_read_counters);
@@ -372,9 +409,9 @@ static unsigned int tegra194_get_speed(u32 cpu)
 		if (pos->driver_data != ndiv)
 			continue;
 
-		if (abs(pos->frequency - rate) > 115200) {
-			pr_warn("cpufreq: cpu%d,cur:%u,set:%u,set ndiv:%llu\n",
-				cpu, rate, pos->frequency, ndiv);
+		if (abs(pos->frequency - rate) > MAX_DELTA_KHZ) {
+			pr_warn("cpufreq: cpu%d,cur:%u,set:%u,delta:%d,set ndiv:%llu\n",
+				cpu, rate, pos->frequency, abs(rate - pos->frequency), ndiv);
 		} else {
 			rate = pos->frequency;
 		}
@@ -568,6 +605,7 @@ static const struct tegra_cpufreq_soc tegra194_cpufreq_soc = {
 	.ops = &tegra194_cpufreq_ops,
 	.maxcpus_per_cluster = 2,
 	.num_clusters = 4,
+	.refclk_delta_min = 16000,
 };
 
 static void tegra194_cpufreq_free_resources(void)
@@ -684,7 +722,7 @@ static int tegra194_cpufreq_probe(struct platform_device *pdev)
 
 	soc = of_device_get_match_data(&pdev->dev);
 
-	if (soc->ops && soc->maxcpus_per_cluster && soc->num_clusters) {
+	if (soc->ops && soc->maxcpus_per_cluster && soc->num_clusters && soc->refclk_delta_min) {
 		data->soc = soc;
 	} else {
 		dev_err(&pdev->dev, "soc data missing\n");
-- 
2.17.1