Skip to contentSkip to conversation

cpufreq: tegra194: get consistent cpuinfo_cur_freq

PATCH Applied Sumit Gupta 2 participants 5 messages Oct 14, 2020 – Oct 28, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 cpufreq: tegra194: get consistent cpuinfo_cur_freq

Frequency returned by 'cpuinfo_cur_freq' using counters is not fixed and keeps changing slightly. This change returns a consistent value from freq_table. If the reconstructed frequency has acceptable delta from the last written value, then return the frequency corresponding to the last written ndiv value from freq_table. Otherwise, print a warning and return the reconstructed freq.

---

Sending only this patch as other patch not required after the change to convert 'pr_warn' to 'pr_info' in cpufreq core for unlisted freq. Changelog
v1[2] -> v3:
- Removed unwanted checks for cpu_online and max cluster number
- Used WARN_ON_ONCE to avoid print flooding.

v1[1] -> v2:
- Minor changes to improve comments and reduce debug prints.
- Get freq table from cluster specific data instead of policy.

[2] https://marc.info/?l=linux-tegra&m=160216218511280&w=2
[1] https://marc.info/?l=linux-arm-kernel&m=160028821117535&w=2

 drivers/cpufreq/tegra194-cpufreq.c | 62 ++++++++++++++++++++++++++++++++------
 1 file changed, 53 insertions(+), 9 deletions(-)
drivers/cpufreq/tegra194-cpufreq.c +53 -9
index e1d931c..7901587 100644
@@ -180,9 +180,61 @@ static unsigned int tegra194_get_speed_common(u32 cpu, u32 delay)
 	return (rate_mhz * KHZ); /* in KHz */
 }
 
+static void get_cpu_ndiv(void *ndiv)
+{
+	u64 ndiv_val;
+
+	asm volatile("mrs %0, s3_0_c15_c0_4" : "=r" (ndiv_val) : );
+
+	*(u64 *)ndiv = ndiv_val;
+}
+
+static void set_cpu_ndiv(void *data)
+{
+	struct cpufreq_frequency_table *tbl = data;
+	u64 ndiv_val = (u64)tbl->driver_data;
+
+	asm volatile("msr s3_0_c15_c0_4, %0" : : "r" (ndiv_val));
+}
+
 static unsigned int tegra194_get_speed(u32 cpu)
 {
-	return tegra194_get_speed_common(cpu, US_DELAY);
+	struct tegra194_cpufreq_data *data = cpufreq_get_driver_data();
+	struct cpufreq_frequency_table *pos;
+	unsigned int rate;
+	u64 ndiv;
+	int ret;
+	u32 cl;
+
+	smp_call_function_single(cpu, get_cpu_cluster, &cl, true);
+
+	/* reconstruct actual cpu freq using counters */
+	rate = tegra194_get_speed_common(cpu, US_DELAY);
+
+	/* get last written ndiv value */
+	ret = smp_call_function_single(cpu, get_cpu_ndiv, &ndiv, true);
+	if (WARN_ON_ONCE(ret))
+		return rate;
+
+	/*
+	 * If the reconstructed frequency has acceptable delta from
+	 * the last written value, then return freq corresponding
+	 * to the last written ndiv value from freq_table. This is
+	 * done to return consistent value.
+	 */
+	cpufreq_for_each_valid_entry(pos, data->tables[cl]) {
+		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);
+		} else {
+			rate = pos->frequency;
+		}
+		break;
+	}
+	return rate;
 }
 
 static int tegra194_cpufreq_init(struct cpufreq_policy *policy)
@@ -209,14 +261,6 @@ static int tegra194_cpufreq_init(struct cpufreq_policy *policy)
 	return 0;
 }
 
-static void set_cpu_ndiv(void *data)
-{
-	struct cpufreq_frequency_table *tbl = data;
-	u64 ndiv_val = (u64)tbl->driver_data;
-
-	asm volatile("msr s3_0_c15_c0_4, %0" : : "r" (ndiv_val));
-}
-
 static int tegra194_cpufreq_set_target(struct cpufreq_policy *policy,
 				       unsigned int index)
 {
Signature · 1 line
2.7.4
#1 lore raw

Ping.

Quoted 113 lines

Frequency returned by 'cpuinfo_cur_freq' using counters is not fixed and keeps changing slightly. This change returns a consistent value from freq_table. If the reconstructed frequency has acceptable delta from the last written value, then return the frequency corresponding to the last written ndiv value from freq_table. Otherwise, print a warning and return the reconstructed freq.

---

Sending only this patch as other patch not required after the change to convert 'pr_warn' to 'pr_info' in cpufreq core for unlisted freq. Changelog
v1[2] -> v3:
- Removed unwanted checks for cpu_online and max cluster number
- Used WARN_ON_ONCE to avoid print flooding.

v1[1] -> v2:
- Minor changes to improve comments and reduce debug prints.
- Get freq table from cluster specific data instead of policy.

[2] https://marc.info/?l=linux-tegra&m=160216218511280&w=2
[1] https://marc.info/?l=linux-arm-kernel&m=160028821117535&w=2

  drivers/cpufreq/tegra194-cpufreq.c | 62 ++++++++++++++++++++++++++++++++------
  1 file changed, 53 insertions(+), 9 deletions(-)
drivers/cpufreq/tegra194-cpufreq.c +53 -10
index e1d931c..7901587 100644
@@ -180,9 +180,61 @@ static unsigned int tegra194_get_speed_common(u32 cpu, u32 delay)
         return (rate_mhz * KHZ); /* in KHz */
  }
 
+static void get_cpu_ndiv(void *ndiv)
+{
+       u64 ndiv_val;
+
+       asm volatile("mrs %0, s3_0_c15_c0_4" : "=r" (ndiv_val) : );
+
+       *(u64 *)ndiv = ndiv_val;
+}
+
+static void set_cpu_ndiv(void *data)
+{
+       struct cpufreq_frequency_table *tbl = data;
+       u64 ndiv_val = (u64)tbl->driver_data;
+
+       asm volatile("msr s3_0_c15_c0_4, %0" : : "r" (ndiv_val));
+}
+
  static unsigned int tegra194_get_speed(u32 cpu)
  {
-       return tegra194_get_speed_common(cpu, US_DELAY);
+       struct tegra194_cpufreq_data *data = cpufreq_get_driver_data();
+       struct cpufreq_frequency_table *pos;
+       unsigned int rate;
+       u64 ndiv;
+       int ret;
+       u32 cl;
+
+       smp_call_function_single(cpu, get_cpu_cluster, &cl, true);
+
+       /* reconstruct actual cpu freq using counters */
+       rate = tegra194_get_speed_common(cpu, US_DELAY);
+
+       /* get last written ndiv value */
+       ret = smp_call_function_single(cpu, get_cpu_ndiv, &ndiv, true);
+       if (WARN_ON_ONCE(ret))
+               return rate;
+
+       /*
+        * If the reconstructed frequency has acceptable delta from
+        * the last written value, then return freq corresponding
+        * to the last written ndiv value from freq_table. This is
+        * done to return consistent value.
+        */
+       cpufreq_for_each_valid_entry(pos, data->tables[cl]) {
+               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);
+               } else {
+                       rate = pos->frequency;
+               }
+               break;
+       }
+       return rate;
  }
 
  static int tegra194_cpufreq_init(struct cpufreq_policy *policy)
@@ -209,14 +261,6 @@ static int tegra194_cpufreq_init(struct cpufreq_policy *policy)
         return 0;
  }
 
-static void set_cpu_ndiv(void *data)
-{
-       struct cpufreq_frequency_table *tbl = data;
-       u64 ndiv_val = (u64)tbl->driver_data;
-
-       asm volatile("msr s3_0_c15_c0_4, %0" : : "r" (ndiv_val));
-}
-
  static int tegra194_cpufreq_set_target(struct cpufreq_policy *policy,
                                        unsigned int index)
  {
--

2.7.4

maintainer
#2 lore raw
Quoted 1 line · Sumit Gupta

Ping.

I was waiting for 5.10-rc1 to be released before I can start applying stuff for 5.11. Now that it is released, I will apply this.

Signature · 1 line
viresh
#3 lore raw

Thank you.

Quoted 12 lines · Viresh Kumar

External email: Use caution opening links or attachments

Quoted 1 line · Sumit Gupta

Ping.

I was waiting for 5.10-rc1 to be released before I can start applying stuff for 5.11. Now that it is released, I will apply this.

--
viresh

maintainer Applied
#4 lore raw
In reply to Sumit Gupta
Quoted 21 lines · Sumit Gupta

Frequency returned by 'cpuinfo_cur_freq' using counters is not fixed and keeps changing slightly. This change returns a consistent value from freq_table. If the reconstructed frequency has acceptable delta from the last written value, then return the frequency corresponding to the last written ndiv value from freq_table. Otherwise, print a warning and return the reconstructed freq.

---

Sending only this patch as other patch not required after the change to convert 'pr_warn' to 'pr_info' in cpufreq core for unlisted freq. Changelog
v1[2] -> v3:
- Removed unwanted checks for cpu_online and max cluster number
- Used WARN_ON_ONCE to avoid print flooding.

v1[1] -> v2:
- Minor changes to improve comments and reduce debug prints.
- Get freq table from cluster specific data instead of policy.

Quoted 5 lines

[2] https://marc.info/?l=linux-tegra&m=160216218511280&w=2
[1] https://marc.info/?l=linux-arm-kernel&m=160028821117535&w=2

 drivers/cpufreq/tegra194-cpufreq.c | 62 ++++++++++++++++++++++++++++++++------
 1 file changed, 53 insertions(+), 9 deletions(-)

Applied. Thanks.

Signature · 6 lines
viresh

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel