Wrong temperature reading

[warped-rudi]

On one of my CuBox-i4p (early model), the CPU temperature exposed via hwmon is bogus (i.e. lower than the ambient). I'd guess that it's off by about 20°C. This does not happen on a second CuBox-i4p (newer model) as well as on HB-i2ex and C1-solo

[linux4kix]

What are the temperatures it is reporting? Does it move up and down
properly when in use. Please use something like cpuburn to increase the
temp.

On Sat, Dec 20, 2014 at 10:33 AM, Rudi Ihle notifications@github.com
wrote:

On one of my CuBox-i4p (early model), the CPU temperature exposed via
hwmon is bogus (i.e. lower than the ambient). I'd guess that it's off by
about 20°C. This does not happen on a second CuBox-i4p (newer model) as
well as on HB-i2ex and C1-solo

—
Reply to this email directly or view it on GitHub
#25.

[warped-rudi]

Yes, it does move up and down. I simulated the same load situation on both CuBoxes and the difference pretty much exactly 20°K.

[warped-rudi]

This seems to be a problem of this particular box. I reverted to the kernel to a version before the thermal changes and the behavior is the same. Probably I didn't pay attention to this in the past. I even got negative temperature readings right after a cold boot. So either the sensor calibration data is bad, the sensor hardware is broken or the clocks are wrong...

[susisstrolch]

Same problem here - with an OE5RC3...
Old CBi4pro (Feb '14) - approx 33° when viewing HD content
New CBi4pro (Dec '14) - approx 64° - same content.
Seems we should chease Rabeeh about...

[linux4kix]

Are the Older CBi's dev units or purchased units?

On Mon, Dec 22, 2014 at 12:02 PM, susisstrolch notifications@github.com
wrote:

Same problem here - with an OE5RC3...
Old CBi4pro (Feb '14) - approx 33° when viewing HD content
New CBi4pro (Dec '14) - approx 64° - same content.
Seems we should chease Rabeeh about...

—
Reply to this email directly or view it on GitHub
#25 (comment)
.

[susisstrolch]

Both CBis are purchased ones...

Just did some testing:
Idling OE5RC5, HDMI, 1920x1080, 50Hz
old: core: 44°, case: 34°, clock 396000
new: core: 64°, case: 42°, clock 396000

case measured with PT100 probe.

dtb's are identical, kernel is 3.14.25

On 12/22/2014 12:56 PM, Jon Nettleton wrote:

Are the Older CBi's dev units or purchased units?

On Mon, Dec 22, 2014 at 12:02 PM, susisstrolch notifications@github.com
wrote:

Same problem here - with an OE5RC3...
Old CBi4pro (Feb '14) - approx 33° when viewing HD content
New CBi4pro (Dec '14) - approx 64° - same content.
Seems we should chease Rabeeh about...

—
Reply to this email directly or view it on GitHub

#25 (comment)
.

—
Reply to this email directly or view it on GitHub
#25 (comment).

[warped-rudi]

@linux4kix: Mine is probably a dev unit. At least it was shipped as such in Jan'14.

@susisstrolch; Is there really such a difference in the case temperature or is that a typo?

@susisstrolch; Can you do a 'devmem 0x21bc4e0 32' That should show the sensor calibration data. I don't have mine handy right now, but I'd like to compare. If they are the same, there is a chance that they are wrong...

[linux4kix]

My guess is that the 25c calibration fuse is wrong.

On Mon, Dec 22, 2014 at 3:50 PM, Rudi Ihle notifications@github.com wrote:

@linux4kix https://github.com/linux4kix: Mine is probably a dev unit.
At least it was shipped as such in Jan'14.

@susisstrolch https://github.com/susisstrolch; I there really such a
difference in the case temperature or is that a typo?

@susisstrolch https://github.com/susisstrolch; Can you do a 'devmem
0x21bc4e0 32' That should show the sensor calibration data. I don't have
mine handy right now, but I'd like to compare. If they are the same, there
is a chance that they are wrong...

—
Reply to this email directly or view it on GitHub
#25 (comment)
.

[susisstrolch]

Yes, thats the real temp diff. Checked twice...

Calibration data is identical on both boxes:

CuBox-i4pro:~ # devmem 0x21bc4e0 32
0x5624D869

Cubox-I4:~ # ./devmem 0x21bc4e0 32
0x5694D969

On 12/22/2014 03:50 PM, Rudi Ihle wrote:

@linux4kix https://github.com/linux4kix: Mine is probably a dev
unit. At least it was shipped as such in Jan'14.

@susisstrolch https://github.com/susisstrolch; I there really such a
difference in the case temperature or is that a typo?

@susisstrolch https://github.com/susisstrolch; Can you do a 'devmem
0x21bc4e0 32' That should show the sensor calibration data. I don't
have mine handy right now, but I'd like to compare. If they are the
same, there is a chance that they are wrong...

—
Reply to this email directly or view it on GitHub
#25 (comment).

[warped-rudi]

@linux4kix I think the problem is with 2b1601b . To me it looks like the new 'universal formula' should only be applied to parts that were calibrated according to this. Given the time of the commit, I suspect this applies to chips manufactured after Feb'14. I did a manual calculation with the calibration data of my 'bad' unit and got exactly the difference I observed. Now, the question is how to auto-detect that. The commit message says: 'there will be no hot point calibration data in fuse map from now on'. However, the 'good' units do have hot point data as well! Have not (yet) checked if they are valid and if the old formula will still work.

@susisstrolch 0x5624D869 != 0x5694D969, still puzzled about the temperature difference also 64°C looks a bit high for an idle device

[linux4kix]

Rudi, if you revert that patch does temperature reading work properly for
both units? Perhaps we need to test if there is hot point calibration data
and then only use the new formula if there isn't. We should also report
this to fsl as this is a patch pushed to upstream as well and could damage
older chips if pushed hard enough.

64C looks hot for idle, but it may be XBMC running idle which is far from
idle. In those circumstances 64C looks just about right to me.

On Tue, Dec 23, 2014 at 1:00 PM, Rudi Ihle notifications@github.com wrote:

@linux4kix https://github.com/linux4kix I think the problem is with
2b1601b
2b1601b
. To me it looks like the new 'universal formula' should only be applied to
parts that were calibrated according to this. Given the time of the commit,
I suspect this applies to chips manufactured after Feb'14. I did a manual
calculation with the calibration data of my 'bad' unit and got exactly the
difference I observed. Now, the question is how to auto-detect that. The
commit message says: 'there will be no hot point calibration data in fuse
map from now on'. However, the 'good' units do have hot point data as well!
Have not (yet) checked if they are valid and if the old formula will still
work.

@susisstrolch https://github.com/susisstrolch 0x5624D869 != 0x5694D969,
still puzzled about the temperature difference also 64°C looks a bit high
for an idle device

—
Reply to this email directly or view it on GitHub
#25 (comment)
.

[warped-rudi]

Will test when I'm at home this evening.

[linux4kix]

great thanks.

On Tue, Dec 23, 2014 at 1:15 PM, Rudi Ihle notifications@github.com wrote:

Will test when I'm at home this evening.

—
Reply to this email directly or view it on GitHub
#25 (comment)
.

[susisstrolch]

Ooops - sloppy pattern matching w/o glasses...

With KODI stopped I get the following values:

CBi Old: 25,00°C @396MHz, i.MX6Q, silicon rev 1.2
CBi New: 40,68°C @396MHz, i.MX6Q, silicon rev 1.5

On 12/23/2014 01:00 PM, Rudi Ihle wrote:

@linux4kix https://github.com/linux4kix I think the problem is with
2b1601b
2b1601b
. To me it looks like the new 'universal formula' should only be
applied to parts that were calibrated according to this. Given the
time of the commit, I suspect this applies to chips manufactured after
Feb'14. I did a manual calculation with the calibration data of my
'bad' unit and got exactly the difference I observed. Now, the
question is how to auto-detect that. The commit message says: 'there
will be no hot point calibration data in fuse map from now on'.
However, the 'good' units do have hot point data as well! Have not
(yet) checked if they are valid and if the old formula will still work.

@susisstrolch https://github.com/susisstrolch 0x5624D869 !=
0x5694D969, still puzzled about the temperature difference also 64°C
looks a bit high for an idle device

—
Reply to this email directly or view it on GitHub
#25 (comment).

[linux4kix]

Okay those numbers look better. If reverting the patch fixes the older
silicon rev then we may be able to use that as the identifier for the
algorithm used.
On Dec 23, 2014 2:14 PM, "susisstrolch" notifications@github.com wrote:

Ooops - sloppy pattern matching w/o glasses...

With KODI stopped I get the following values:

CBi Old: 25,00°C @396MHz, i.MX6Q, silicon rev 1.2
CBi New: 40,68°C @396MHz, i.MX6Q, silicon rev 1.5

On 12/23/2014 01:00 PM, Rudi Ihle wrote:

@linux4kix https://github.com/linux4kix I think the problem is with
2b1601b
<
2b1601b6976a838029fd7695dabab189358acbc0>

. To me it looks like the new 'universal formula' should only be
applied to parts that were calibrated according to this. Given the
time of the commit, I suspect this applies to chips manufactured after
Feb'14. I did a manual calculation with the calibration data of my
'bad' unit and got exactly the difference I observed. Now, the
question is how to auto-detect that. The commit message says: 'there
will be no hot point calibration data in fuse map from now on'.
However, the 'good' units do have hot point data as well! Have not
(yet) checked if they are valid and if the old formula will still work.

@susisstrolch https://github.com/susisstrolch 0x5624D869 !=
0x5694D969, still puzzled about the temperature difference also 64°C
looks a bit high for an idle device

—
Reply to this email directly or view it on GitHub
<
#25 (comment)
.

—
Reply to this email directly or view it on GitHub
#25 (comment)
.