Skip to contentSkip to conversation

hwmon: (coretemp) Do not return -EAGAIN for low temperatures

PATCH In review Guenter Roeck 2 participants 5 messages Dec 7, 2013 – Dec 21, 2013

Subsystems (2)
lore
Share & tools
lore

Guenter Roeck

author maintainer
#0 lore raw

PATCH 1/1 hwmon: (coretemp) Do not return -EAGAIN for low temperatures

Some Intel CPUs do not set the 'valid' bit in IA32_THERM_STATUS if the temperature is too low to be measured. This condition will not change until the CPU is hot enough for its temperature to be measured. Returning an error in such conditions is not very useful. Drop checking the valid bit and just return the reported temperature instead.

---
I don't think we ever closed on this. Giving it a shot.

 drivers/hwmon/coretemp.c |   17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)
drivers/hwmon/coretemp.c +9 -8
index 310ce19..6e0579e 100644
@@ -177,18 +177,19 @@ static ssize_t show_temp(struct device *dev,
 	/* Check whether the time interval has elapsed */
 	if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
 		rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
-		tdata->valid = 0;
-		/* Check whether the data is valid */
-		if (eax & 0x80000000) {
-			tdata->temp = tdata->tjmax -
-					((eax >> 16) & 0x7f) * 1000;
-			tdata->valid = 1;
-		}
+		/*
+		 * Ignore the valid bit. In all observed cases the register
+		 * value is either low or zero if the valid bit is 0.
+		 * Return it instead of reporting an error which doesn't
+		 * really help at all.
+		 */
+		tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
+		tdata->valid = 1;
 		tdata->last_updated = jiffies;
 	}
 
 	mutex_unlock(&tdata->update_lock);
-	return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
+	return sprintf(buf, "%d\n", tdata->temp);
 }
 
 struct tjmax_pci {
Signature · 7 lines
1.7.9.7


_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
author maintainer
#1 lore raw
Quoted 10 lines · Guenter Roeck

Some Intel CPUs do not set the 'valid' bit in IA32_THERM_STATUS if the temperature is too low to be measured. This condition will not change until the CPU is hot enough for its temperature to be measured. Returning an error in such conditions is not very useful. Drop checking the valid bit and just return the reported temperature instead.

---
I don't think we ever closed on this. Giving it a shot.

Comments, anyone ?

Guenter

Quoted 26 lines
  drivers/hwmon/coretemp.c |   17 +++++++++--------
  1 file changed, 9 insertions(+), 8 deletions(-)
drivers/hwmon/coretemp.c +7 -7
index 310ce19..6e0579e 100644
@@ -177,18 +177,19 @@ static ssize_t show_temp(struct device *dev,
  	/* Check whether the time interval has elapsed */
  	if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
  		rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
-		tdata->valid = 0;
-		/* Check whether the data is valid */
-		if (eax & 0x80000000) {
-			tdata->temp = tdata->tjmax -
-					((eax >> 16) & 0x7f) * 1000;
-			tdata->valid = 1;
-		}
+		/*
+		 * Ignore the valid bit. In all observed cases the register
+		 * value is either low or zero if the valid bit is 0.
+		 * Return it instead of reporting an error which doesn't
+		 * really help at all.
+		 */
+		tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;

drivers/hwmon/coretemp.c static ssize_t show_temp(struct device *dev, · line 195

Quoted 10 lines
drivers/hwmon/coretemp.c +2 -1
+		tdata->valid = 1;
  		tdata->last_updated = jiffies;
  	}
 
  	mutex_unlock(&tdata->update_lock);
-	return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
+	return sprintf(buf, "%d\n", tdata->temp);
  }
 
  struct tjmax_pci {

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

author maintainer
#2 lore raw
In reply to Guenter Roeck
Quoted 10 lines · Guenter Roeck

Some Intel CPUs do not set the 'valid' bit in IA32_THERM_STATUS if the temperature is too low to be measured. This condition will not change until the CPU is hot enough for its temperature to be measured. Returning an error in such conditions is not very useful. Drop checking the valid bit and just return the reported temperature instead.

---
I don't think we ever closed on this. Giving it a shot.

No feedback. This will go into 3.14 unless there are objections.

Guenter

Quoted 39 lines
 drivers/hwmon/coretemp.c |   17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)
drivers/hwmon/coretemp.c +9 -8
index 310ce19..6e0579e 100644
@@ -177,18 +177,19 @@ static ssize_t show_temp(struct device *dev,
 	/* Check whether the time interval has elapsed */
 	if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
 		rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
-		tdata->valid = 0;
-		/* Check whether the data is valid */
-		if (eax & 0x80000000) {
-			tdata->temp = tdata->tjmax -
-					((eax >> 16) & 0x7f) * 1000;
-			tdata->valid = 1;
-		}
+		/*
+		 * Ignore the valid bit. In all observed cases the register
+		 * value is either low or zero if the valid bit is 0.
+		 * Return it instead of reporting an error which doesn't
+		 * really help at all.
+		 */
+		tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
+		tdata->valid = 1;
 		tdata->last_updated = jiffies;
 	}
 
 	mutex_unlock(&tdata->update_lock);
-	return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
+	return sprintf(buf, "%d\n", tdata->temp);
 }
 
 struct tjmax_pci {
Signature · 1 line
1.7.9.7

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Reviewed-by on v1 1/1
#3 lore raw
In reply to Guenter Roeck
Quoted 7 lines · Guenter Roeck
Quoted 6 lines · Guenter Roeck

Some Intel CPUs do not set the 'valid' bit in IA32_THERM_STATUS if the temperature is too low to be measured. This condition will not change until the CPU is hot enough for its temperature to be measured. Returning an error in such conditions is not very useful. Drop checking the valid bit and just return the reported temperature instead.

Quoted 5 lines
Quoted 4 lines

---
I don't think we ever closed on this. Giving it a shot.

No feedback. This will go into 3.14 unless there are objections.

I have no objection, I just don't want this change to go to stable, I believe it needs a lot of testing on a broad range of hardware before we can even think of it.

On a related note, I think we already noticed that different CPU models behave differently in low temperature ranges. As an additional data point, I noticed when staring at sensord graphs that my wife's Core2 Duo E8400 does clamp at 40°C (which is 60°C below Tjmax). It never ever reports values below this. Somehow it is a saner implementation than clearing the valid bit. Too bad we have no way to represent this in sysfs. Would it make sense to define for example tempX_floor (can't remember if a better name was ever proposed) so that sensors can report "< 40°C" instead of "40°C" in this case? The problem is that I don't think these clamp values can be read from a register, I'm not even sure if they are documented, so I'm not sure how useful this would be in practice...

Signature · 6 lines
Jean Delvare

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
author maintainer
#4 lore raw
In reply to Guenter Roeck
Quoted 12 lines · Jean Delvare
Quoted 11 lines · Guenter Roeck
Quoted 10 lines · Guenter Roeck

Some Intel CPUs do not set the 'valid' bit in IA32_THERM_STATUS if the temperature is too low to be measured. This condition will not change until the CPU is hot enough for its temperature to be measured. Returning an error in such conditions is not very useful. Drop checking the valid bit and just return the reported temperature instead.

---
I don't think we ever closed on this. Giving it a shot.

Quoted 6 lines
Quoted 1 line

No feedback. This will go into 3.14 unless there are objections.

I have no objection, I just don't want this change to go to stable, I believe it needs a lot of testing on a broad range of hardware before we can even think of it.

Agreed.

Quoted 15 lines

On a related note, I think we already noticed that different CPU models behave differently in low temperature ranges. As an additional data point, I noticed when staring at sensord graphs that my wife's Core2 Duo E8400 does clamp at 40°C (which is 60°C below Tjmax). It never ever reports values below this. Somehow it is a saner implementation than clearing the valid bit. Too bad we have no way to represent this in sysfs. Would it make sense to define for example tempX_floor (can't remember if a better name was ever proposed) so that sensors can report "< 40°C" instead of "40°C" in this case? The problem is that I don't think these clamp values can be read from a register, I'm not even sure if they are documented, so I'm not sure how useful this would be in practice...

Question is really if we would ever have a chance to determine the value to report as floor. I suspect the answer is no.

Thanks,
Guenter

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors