Skip to contentSkip to conversation

Fix RPM calculation in pwm-fan

PATCH Superseded Paul Barker 2 participants 7 messages Nov 10, 2020 Hardware monitoring

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v3 cover Fix RPM calculation in pwm-fan

Fix inaccuracy and inefficiency when calculating fan RPM in the pwm-fan driver.

These changes can also be pulled from:

https://gitlab.com/pbarker.dev/staging/linux.git
tag: for-hwmon/pwm-fan-fix-rpm-v3_2020-11-10

Changes from v2:

* Handle the case where delta=0 to avoid potential divide-by-zero exceptions.

* Store a device pointer in pwm_fan_ctx so we can print an error message.

Paul Barker (2):
hwmon: pwm-fan: Store device pointer in pwm_fan_ctx
hwmon: pwm-fan: Fix RPM calculation

 drivers/hwmon/pwm-fan.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

base-commit: f8394f232b1eab649ce2df5c5f15b0e528c92091

Signature · 1 line
2.29.2
#1 lore raw

PATCH v3 1/2 hwmon: pwm-fan: Store device pointer in pwm_fan_ctx

This allows us to use dev_err() and friends from functions which only have a pwm_fan_ctx pointer.

---
 drivers/hwmon/pwm-fan.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/hwmon/pwm-fan.c +2 -0
index bdba2143021a..edc0453be25a 100644
@@ -26,6 +26,7 @@ struct pwm_fan_ctx {
 	struct mutex lock;
 	struct pwm_device *pwm;
 	struct regulator *reg_en;
+	struct device *dev;
 
 	int irq;
 	atomic_t pulses;
@@ -290,6 +291,7 @@ static int pwm_fan_probe(struct platform_device *pdev)
 	if (!ctx)
 		return -ENOMEM;
 
+	ctx->dev = dev;
 	mutex_init(&ctx->lock);
 
 	ctx->pwm = devm_of_pwm_get(dev, dev->of_node, NULL);
Signature · 1 line
2.29.2
#2 lore raw

PATCH v3 2/2 hwmon: pwm-fan: Fix RPM calculation

In reply to Paul Barker

To convert the number of pulses counted into an RPM estimation, we need to divide by the width of our measurement interval instead of multiplying by it. If the width of the measurement interval is zero we bail out instead of dividing by it.

We also don't need to do 64-bit division, with 32-bits we can handle a fan running at over 4 million RPM.

---
 drivers/hwmon/pwm-fan.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/hwmon/pwm-fan.c +11 -6
index edc0453be25a..24cfed4d625e 100644
@@ -55,14 +55,19 @@ static irqreturn_t pulse_handler(int irq, void *dev_id)
 static void sample_timer(struct timer_list *t)
 {
 	struct pwm_fan_ctx *ctx = from_timer(ctx, t, rpm_timer);
+	unsigned int delta = ktime_ms_delta(ktime_get(), ctx->sample_start);
 	int pulses;
-	u64 tmp;
 
-	pulses = atomic_read(&ctx->pulses);
-	atomic_sub(pulses, &ctx->pulses);
-	tmp = (u64)pulses * ktime_ms_delta(ktime_get(), ctx->sample_start) * 60;
-	do_div(tmp, ctx->pulses_per_revolution * 1000);
-	ctx->rpm = tmp;
+	if (delta) {
+		pulses = atomic_read(&ctx->pulses);
+		atomic_sub(pulses, &ctx->pulses);
+		ctx->rpm = (unsigned int)(pulses * 1000 * 60) /
+			(ctx->pulses_per_revolution * delta);
+	} else {
+		dev_err(ctx->dev,
+			"Cannot determine fan RPM as time delta is zero\n");
+		ctx->rpm = 0;
+	}
 
 	ctx->sample_start = ktime_get();
 	mod_timer(&ctx->rpm_timer, jiffies + HZ);
Signature · 1 line
2.29.2
maintainer
#3 lore raw

Re: [PATCH v3 2/2] hwmon: pwm-fan: Fix RPM calculation

Quoted 29 lines · Paul Barker

To convert the number of pulses counted into an RPM estimation, we need to divide by the width of our measurement interval instead of multiplying by it. If the width of the measurement interval is zero we bail out instead of dividing by it.

We also don't need to do 64-bit division, with 32-bits we can handle a fan running at over 4 million RPM.

---
 drivers/hwmon/pwm-fan.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/hwmon/pwm-fan.c +1 -5
index edc0453be25a..24cfed4d625e 100644
@@ -55,14 +55,19 @@ static irqreturn_t pulse_handler(int irq, void *dev_id)
 static void sample_timer(struct timer_list *t)
 {
 	struct pwm_fan_ctx *ctx = from_timer(ctx, t, rpm_timer);
+	unsigned int delta = ktime_ms_delta(ktime_get(), ctx->sample_start);
 	int pulses;
-	u64 tmp;
 
-	pulses = atomic_read(&ctx->pulses);
-	atomic_sub(pulses, &ctx->pulses);
-	tmp = (u64)pulses * ktime_ms_delta(ktime_get(), ctx->sample_start) * 60;
-	do_div(tmp, ctx->pulses_per_revolution * 1000);

drivers/hwmon/pwm-fan.c static irqreturn_t pulse_handler(int irq, void *dev_id) · line 69

Quoted 10 lines
drivers/hwmon/pwm-fan.c +9 -1
-	ctx->rpm = tmp;
+	if (delta) {
+		pulses = atomic_read(&ctx->pulses);
+		atomic_sub(pulses, &ctx->pulses);
+		ctx->rpm = (unsigned int)(pulses * 1000 * 60) /
+			(ctx->pulses_per_revolution * delta);
+	} else {
+		dev_err(ctx->dev,
+			"Cannot determine fan RPM as time delta is zero\n");
+		ctx->rpm = 0;

I don't think that warrants an error message. At best it should be a debug message, but even that seems not worth it. I would suggest to not update rpm if that happens. After all, it is pretty much a theoretic case.

Guenter

Quoted 7 lines
drivers/hwmon/pwm-fan.c +1 -0
+	}
 
 	ctx->sample_start = ktime_get();
 	mod_timer(&ctx->rpm_timer, jiffies + HZ);
Signature · 1 line
2.29.2
#4 lore raw

Re: [PATCH v3 2/2] hwmon: pwm-fan: Fix RPM calculation

Quoted 37 lines · Guenter Roeck
Quoted 35 lines · Paul Barker

To convert the number of pulses counted into an RPM estimation, we need to divide by the width of our measurement interval instead of multiplying by it. If the width of the measurement interval is zero we bail out instead of dividing by it.

We also don't need to do 64-bit division, with 32-bits we can handle a fan running at over 4 million RPM.

---
 drivers/hwmon/pwm-fan.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/hwmon/pwm-fan.c +6 -6
index edc0453be25a..24cfed4d625e 100644
@@ -55,14 +55,19 @@ static irqreturn_t pulse_handler(int irq, void *dev_id)
 static void sample_timer(struct timer_list *t)
 {
      struct pwm_fan_ctx *ctx = from_timer(ctx, t, rpm_timer);
+     unsigned int delta = ktime_ms_delta(ktime_get(), ctx->sample_start);
      int pulses;
-     u64 tmp;
 
-     pulses = atomic_read(&ctx->pulses);
-     atomic_sub(pulses, &ctx->pulses);
-     tmp = (u64)pulses * ktime_ms_delta(ktime_get(), ctx->sample_start) * 60;
-     do_div(tmp, ctx->pulses_per_revolution * 1000);
-     ctx->rpm = tmp;
+     if (delta) {
+             pulses = atomic_read(&ctx->pulses);
+             atomic_sub(pulses, &ctx->pulses);
+             ctx->rpm = (unsigned int)(pulses * 1000 * 60) /
+                     (ctx->pulses_per_revolution * delta);
Quoted 8 lines

drivers/hwmon/pwm-fan.c static irqreturn_t pulse_handler(int irq, void *dev_id) · line 69

Quoted 4 lines
drivers/hwmon/pwm-fan.c +4 -0
+     } else {
+             dev_err(ctx->dev,
+                     "Cannot determine fan RPM as time delta is zero\n");
+             ctx->rpm = 0;

I don't think that warrants an error message. At best it should be a debug message, but even that seems not worth it. I would suggest to not update rpm if that happens. After all, it is pretty much a theoretic case.

My thought process was that setting an RPM value of zero would be confusing - it could be caused due to fan failure or due to this (theoretical) error. I'm happy to drop the error message though - is the patch acceptable other than that?

Signature · 2 lines
Paul Barker
Konsulko Group
maintainer
#5 lore raw

Re: [PATCH v3 2/2] hwmon: pwm-fan: Fix RPM calculation

Quoted 47 lines · Paul Barker
Quoted 37 lines · Guenter Roeck
Quoted 35 lines · Paul Barker

To convert the number of pulses counted into an RPM estimation, we need to divide by the width of our measurement interval instead of multiplying by it. If the width of the measurement interval is zero we bail out instead of dividing by it.

We also don't need to do 64-bit division, with 32-bits we can handle a fan running at over 4 million RPM.

---
 drivers/hwmon/pwm-fan.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/hwmon/pwm-fan.c +6 -6
index edc0453be25a..24cfed4d625e 100644
@@ -55,14 +55,19 @@ static irqreturn_t pulse_handler(int irq, void *dev_id)
 static void sample_timer(struct timer_list *t)
 {
      struct pwm_fan_ctx *ctx = from_timer(ctx, t, rpm_timer);
+     unsigned int delta = ktime_ms_delta(ktime_get(), ctx->sample_start);
      int pulses;
-     u64 tmp;
 
-     pulses = atomic_read(&ctx->pulses);
-     atomic_sub(pulses, &ctx->pulses);
-     tmp = (u64)pulses * ktime_ms_delta(ktime_get(), ctx->sample_start) * 60;
-     do_div(tmp, ctx->pulses_per_revolution * 1000);
-     ctx->rpm = tmp;
+     if (delta) {
+             pulses = atomic_read(&ctx->pulses);
+             atomic_sub(pulses, &ctx->pulses);
+             ctx->rpm = (unsigned int)(pulses * 1000 * 60) /
+                     (ctx->pulses_per_revolution * delta);
Quoted 8 lines

drivers/hwmon/pwm-fan.c static irqreturn_t pulse_handler(int irq, void *dev_id) · line 69

Quoted 4 lines
drivers/hwmon/pwm-fan.c +4 -0
+     } else {
+             dev_err(ctx->dev,
+                     "Cannot determine fan RPM as time delta is zero\n");
+             ctx->rpm = 0;

I don't think that warrants an error message. At best it should be a debug message, but even that seems not worth it. I would suggest to not update rpm if that happens. After all, it is pretty much a theoretic case.

Quoted 3 lines

My thought process was that setting an RPM value of zero would be confusing - it could be caused due to fan failure or due to this (theoretical) error. I'm happy to drop the error message though - is

Yes, that is why I suggested to keep the old speed in that situation. After all, it _will_ be updated shortly afterwards. Either case, people won't typically look into the kernel log if they see the 0 rpm.

Guenter

Quoted 2 lines

the patch acceptable other than that?

#6 lore raw

Re: [PATCH v3 2/2] hwmon: pwm-fan: Fix RPM calculation

Quoted 49 lines · Guenter Roeck
Quoted 47 lines · Paul Barker
Quoted 45 lines · Guenter Roeck
Quoted 29 lines · Paul Barker

To convert the number of pulses counted into an RPM estimation, we need to divide by the width of our measurement interval instead of multiplying by it. If the width of the measurement interval is zero we bail out instead of dividing by it.

We also don't need to do 64-bit division, with 32-bits we can handle a fan running at over 4 million RPM.

---
 drivers/hwmon/pwm-fan.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)
drivers/hwmon/pwm-fan.c +1 -5
index edc0453be25a..24cfed4d625e 100644
@@ -55,14 +55,19 @@ static irqreturn_t pulse_handler(int irq, void *dev_id)
 static void sample_timer(struct timer_list *t)
 {
      struct pwm_fan_ctx *ctx = from_timer(ctx, t, rpm_timer);
+     unsigned int delta = ktime_ms_delta(ktime_get(), ctx->sample_start);
      int pulses;
-     u64 tmp;
 
-     pulses = atomic_read(&ctx->pulses);
-     atomic_sub(pulses, &ctx->pulses);
-     tmp = (u64)pulses * ktime_ms_delta(ktime_get(), ctx->sample_start) * 60;
-     do_div(tmp, ctx->pulses_per_revolution * 1000);

drivers/hwmon/pwm-fan.c static irqreturn_t pulse_handler(int irq, void *dev_id) · line 69

Quoted 10 lines
drivers/hwmon/pwm-fan.c +9 -1
-     ctx->rpm = tmp;
+     if (delta) {
+             pulses = atomic_read(&ctx->pulses);
+             atomic_sub(pulses, &ctx->pulses);
+             ctx->rpm = (unsigned int)(pulses * 1000 * 60) /
+                     (ctx->pulses_per_revolution * delta);
+     } else {
+             dev_err(ctx->dev,
+                     "Cannot determine fan RPM as time delta is zero\n");
+             ctx->rpm = 0;

I don't think that warrants an error message. At best it should be a debug message, but even that seems not worth it. I would suggest to not update rpm if that happens. After all, it is pretty much a theoretic case.

Quoted 7 lines
Quoted 3 lines

My thought process was that setting an RPM value of zero would be confusing - it could be caused due to fan failure or due to this (theoretical) error. I'm happy to drop the error message though - is

Yes, that is why I suggested to keep the old speed in that situation. After all, it _will_ be updated shortly afterwards. Either case, people won't typically look into the kernel log if they see the 0 rpm.

Ah, I misunderstood your earlier reply then. Yes - this makes sense, clearly I haven't had enough coffee as that solution should have been obvious!

v4 incoming.

Signature · 2 lines
Paul Barker
Konsulko Group