From: Eric Woudstra <redacted>
Cpu-thermal is set to use all frequencies already at 47 degrees.
Using the CPU at 50 for a minute, the CPU has reached 48 degrees, is
throttled back to lowest setting, making the mt7622 terrribly slow.
Even at this low speed, the CPU does not cool down lower then 47 so
the CPU is stuck at lowest possible frequency until it shut down and
stays off for 15 minutes.
cooling-device = <&cpu0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>,
<&cpu1 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
This should not be set al every cooling map. It should only be set at
the highest cooling map. Same as in the example:
https://www.kernel.org/doc/Documentation/devicetree/bindings/
thermal/thermal.txt line 272
But then without the fan and added a third map.
Now temperature will be regulated at 87 degrees celcius. At temperatures
lower then 87, all frequencies can be used.
Also see the post:
http://forum.banana-pi.org/t/bpi-r64-only-10-cpu-speed-at-already-48-
degrees-celcius-speed-not-increasing-anymore/12262
Signed-off-by: Eric Woudstra <redacted>
---
arch/arm64/boot/dts/mediatek/mt7622.dtsi | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
From: Daniel Lezcano <hidden> Date: 2021-06-21 18:29:20
On 19/06/2021 14:19, ericwouds@gmail.com wrote:
From: Eric Woudstra <redacted>
Cpu-thermal is set to use all frequencies already at 47 degrees.
Using the CPU at 50 for a minute, the CPU has reached 48 degrees, is
throttled back to lowest setting, making the mt7622 terrribly slow.
Even at this low speed, the CPU does not cool down lower then 47 so
the CPU is stuck at lowest possible frequency until it shut down and
stays off for 15 minutes.
cooling-device = <&cpu0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>,
<&cpu1 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
This should not be set al every cooling map. It should only be set at
the highest cooling map. Same as in the example:
https://www.kernel.org/doc/Documentation/devicetree/bindings/
thermal/thermal.txt line 272
But then without the fan and added a third map.
Now temperature will be regulated at 87 degrees celcius. At temperatures
lower then 87, all frequencies can be used.
47°C is really a too low temperature and this performance drop is normal.
I would not remove the passive mitigation but try by increasing the CPU
temp to 70°C and by changing the active trip point to 80°C. If it works
fine, try 75°C and 85°C.
To test, the thermal killer is dhrystone (one thread per cpu).
With a 75°C passive trip point, the step wise thermal governor, I think
the mitigation will happen smoothly providing better performances, and
probably the fan won't fire.
quoted hunk
Also see the post:
http://forum.banana-pi.org/t/bpi-r64-only-10-cpu-speed-at-already-48-
degrees-celcius-speed-not-increasing-anymore/12262
Signed-off-by: Eric Woudstra <redacted>
---
arch/arm64/boot/dts/mediatek/mt7622.dtsi | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
From: Eric Woudstra <hidden> Date: 2021-06-23 15:35:41
It is only useful to set 1 map with the regulated temperature for cpu frequency throttling. Same as in the kernel document example.
It has no use to set frequency scaling on 2 different temperature trip points, as the lowest one makes sure the higher one(s) are never reached. It can be applied only at 1 trip point. Multiple trip points is only usefully for fan control to make sure the fan is not too noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass the lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
There is no fan... On the bpi r64.
Anyway without throttling at all a kernel build for more then an hour temperature creeps up to 85 degrees.
Get BlueMail for Android
On Jun 21, 2021, 8:29 PM, at 8:29 PM, Daniel Lezcano [off-list ref] wrote:
On 19/06/2021 14:19, ericwouds@gmail.com wrote:
quoted
From: Eric Woudstra <redacted>
Cpu-thermal is set to use all frequencies already at 47 degrees.
Using the CPU at 50 for a minute, the CPU has reached 48 degrees, is
throttled back to lowest setting, making the mt7622 terrribly slow.
Even at this low speed, the CPU does not cool down lower then 47 so
the CPU is stuck at lowest possible frequency until it shut down and
stays off for 15 minutes.
cooling-device = <&cpu0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>,
<&cpu1 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
This should not be set al every cooling map. It should only be set at
the highest cooling map. Same as in the example:
https://www.kernel.org/doc/Documentation/devicetree/bindings/
thermal/thermal.txt line 272
But then without the fan and added a third map.
Now temperature will be regulated at 87 degrees celcius. At
temperatures
quoted
lower then 87, all frequencies can be used.
47°C is really a too low temperature and this performance drop is
normal.
I would not remove the passive mitigation but try by increasing the CPU
temp to 70°C and by changing the active trip point to 80°C. If it works
fine, try 75°C and 85°C.
To test, the thermal killer is dhrystone (one thread per cpu).
With a 75°C passive trip point, the step wise thermal governor, I think
the mitigation will happen smoothly providing better performances, and
probably the fan won't fire.
quoted
Also see the post:
http://forum.banana-pi.org/t/bpi-r64-only-10-cpu-speed-at-already-48-
degrees-celcius-speed-not-increasing-anymore/12262
Signed-off-by: Eric Woudstra <redacted>
---
arch/arm64/boot/dts/mediatek/mt7622.dtsi | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
From: Daniel Lezcano <hidden> Date: 2021-06-23 15:58:37
On 23/06/2021 17:35, Eric Woudstra wrote:
It is only useful to set 1 map with the regulated temperature for cpu
frequency throttling. Same as in the kernel document example.
It has no use to set frequency scaling on 2 different temperature
trip points, as the lowest one makes sure the higher one(s) are never
reached.
I looked more closely the DT and there is a misunderstanding of the
thermal framework in the definition.
There is one trip point with the passive type and the cpu cooling
device, followed by a second trip point with the active type *but* the
same cpu cooling device. That is wrong.
And finally, there is the hot trip point as a third mapping and the same
cooling device.
The hot trip point is only there to notify userspace and let it take an
immediate action to prevent an emergency shutdown when reaching the
critical temperature.
It can be applied only at 1 trip point. Multiple trip points
is only usefully for fan control to make sure the fan is not too
noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass the
lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
Yes, you are right. It should be something like (verbatim copy):
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
From: Eric Woudstra <hidden> Date: 2021-06-23 18:45:17
I choose "hot" before, because 87 degrees seems ok to start frequency throttling. But, yes, it should be passive.
87 is still quite low if I compare this temperature with the wrt3200acm Marvell dual core arm soc. They even went above 100 degrees so I feel for an arm processor inside a router box it is fine to use 87 degrees But maybe someone at Mediatek can give some more details about operating temperatures.
It may be possible to leave the active map in the device tree as some users of the bananapi might choose to install a fan as it is one of the options.
Get BlueMail for Android
On Jun 23, 2021, 5:58 PM, at 5:58 PM, Daniel Lezcano [off-list ref] wrote:
quoted hunk
On 23/06/2021 17:35, Eric Woudstra wrote:
quoted
It is only useful to set 1 map with the regulated temperature for cpu
frequency throttling. Same as in the kernel document example.
It has no use to set frequency scaling on 2 different temperature
trip points, as the lowest one makes sure the higher one(s) are never
reached.
I looked more closely the DT and there is a misunderstanding of the
thermal framework in the definition.
There is one trip point with the passive type and the cpu cooling
device, followed by a second trip point with the active type *but* the
same cpu cooling device. That is wrong.
And finally, there is the hot trip point as a third mapping and the
same
cooling device.
The hot trip point is only there to notify userspace and let it take an
immediate action to prevent an emergency shutdown when reaching the
critical temperature.
quoted
It can be applied only at 1 trip point. Multiple trip points
is only usefully for fan control to make sure the fan is not too
noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass the
lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
Yes, you are right. It should be something like (verbatim copy):
From: Daniel Lezcano <hidden> Date: 2021-06-23 20:08:28
On 23/06/2021 20:43, Eric Woudstra wrote:
I choose "hot" before, because 87 degrees seems ok to start frequency
throttling. But, yes, it should be passive.
87 is still quite low if I compare this temperature with the
wrt3200acm Marvell dual core arm soc. They even went above 100
degrees so I feel for an arm processor inside a router box it is fine
to use 87 degrees But maybe someone at Mediatek can give some more
details about operating temperatures.
Sometimes, the SoC vendor puts a high temperature in the DT just to
export the thermal zone and deal with it from userspace. So putting the
high temp allow the userspace (usually a thermal engine - Android stuff)
to deal with the mitigation without a kernel interaction.
Having more than 100°C could be this kind of setup. Only the operating
temperature from the hardware documentation will tell the safe
temperature for the silicon.
IMO, 77°C is a good compromise until getting the documented temp. 87°C
sounds to me a bit too hot.
It may be possible to leave the active map in the device tree as some
users of the bananapi might choose to install a fan as it is one of
the options.
Get BlueMail for Android
On Jun 23, 2021, 5:58 PM, at 5:58 PM, Daniel Lezcano
[off-list ref] wrote:
quoted
On 23/06/2021 17:35, Eric Woudstra wrote:
quoted
It is only useful to set 1 map with the regulated temperature for
cpu frequency throttling. Same as in the kernel document
example.
It has no use to set frequency scaling on 2 different
temperature trip points, as the lowest one makes sure the higher
one(s) are never reached.
I looked more closely the DT and there is a misunderstanding of
the thermal framework in the definition.
There is one trip point with the passive type and the cpu cooling
device, followed by a second trip point with the active type *but*
the same cpu cooling device. That is wrong.
And finally, there is the hot trip point as a third mapping and
the same cooling device.
The hot trip point is only there to notify userspace and let it
take an immediate action to prevent an emergency shutdown when
reaching the critical temperature.
quoted
It can be applied only at 1 trip point. Multiple trip points is
only usefully for fan control to make sure the fan is not too
noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass
the lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
Yes, you are right. It should be something like (verbatim copy):
I choose "hot" before, because 87 degrees seems ok to start frequency
throttling. But, yes, it should be passive.
87 is still quite low if I compare this temperature with the
wrt3200acm Marvell dual core arm soc. They even went above 100
degrees so I feel for an arm processor inside a router box it is fine
to use 87 degrees But maybe someone at Mediatek can give some more
details about operating temperatures.
Sometimes, the SoC vendor puts a high temperature in the DT just to
export the thermal zone and deal with it from userspace. So putting the
high temp allow the userspace (usually a thermal engine - Android
stuff)
to deal with the mitigation without a kernel interaction.
Having more than 100°C could be this kind of setup. Only the operating
temperature from the hardware documentation will tell the safe
temperature for the silicon.
IMO, 77°C is a good compromise until getting the documented temp. 87°C
sounds to me a bit too hot.
quoted
It may be possible to leave the active map in the device tree as some
users of the bananapi might choose to install a fan as it is one of
the options.
Get BlueMail for Android
On Jun 23, 2021, 5:58 PM, at 5:58 PM, Daniel Lezcano
[off-list ref] wrote:
quoted
On 23/06/2021 17:35, Eric Woudstra wrote:
quoted
It is only useful to set 1 map with the regulated temperature for
cpu frequency throttling. Same as in the kernel document
example.
It has no use to set frequency scaling on 2 different
temperature trip points, as the lowest one makes sure the higher
one(s) are never reached.
I looked more closely the DT and there is a misunderstanding of
the thermal framework in the definition.
There is one trip point with the passive type and the cpu cooling
device, followed by a second trip point with the active type *but*
the same cpu cooling device. That is wrong.
And finally, there is the hot trip point as a third mapping and
the same cooling device.
The hot trip point is only there to notify userspace and let it
take an immediate action to prevent an emergency shutdown when
reaching the critical temperature.
quoted
It can be applied only at 1 trip point. Multiple trip points is
only usefully for fan control to make sure the fan is not too
noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass
the lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
Yes, you are right. It should be something like (verbatim copy):
Found that:
https://download.kamami.pl/p579344-MT7622A_Datasheet_for_BananaPi_Only%281%29.pdf
Chapter 3.3 - Thermal Characteristics
Given the values I suggest:
- Passive - 80°C
- Hot - 90°C
- Critical - 100°C
And passive polling set to 250ms.
It sounds like the sensor is not supporting the interrupt mode yet, so a
big gap is needed with the Tj IMO to give the time to detect the trip
point crossing with the polling.
Get BlueMail for Android
On Jun 23, 2021, 10:08 PM, at 10:08 PM, Daniel Lezcano [off-list ref] wrote:
quoted
On 23/06/2021 20:43, Eric Woudstra wrote:
quoted
I choose "hot" before, because 87 degrees seems ok to start frequency
throttling. But, yes, it should be passive.
87 is still quite low if I compare this temperature with the
wrt3200acm Marvell dual core arm soc. They even went above 100
degrees so I feel for an arm processor inside a router box it is fine
to use 87 degrees But maybe someone at Mediatek can give some more
details about operating temperatures.
Sometimes, the SoC vendor puts a high temperature in the DT just to
export the thermal zone and deal with it from userspace. So putting the
high temp allow the userspace (usually a thermal engine - Android
stuff)
to deal with the mitigation without a kernel interaction.
Having more than 100°C could be this kind of setup. Only the operating
temperature from the hardware documentation will tell the safe
temperature for the silicon.
IMO, 77°C is a good compromise until getting the documented temp. 87°C
sounds to me a bit too hot.
quoted
It may be possible to leave the active map in the device tree as some
users of the bananapi might choose to install a fan as it is one of
the options.
Get BlueMail for Android
On Jun 23, 2021, 5:58 PM, at 5:58 PM, Daniel Lezcano
[off-list ref] wrote:
quoted
On 23/06/2021 17:35, Eric Woudstra wrote:
quoted
It is only useful to set 1 map with the regulated temperature for
cpu frequency throttling. Same as in the kernel document
example.
It has no use to set frequency scaling on 2 different
temperature trip points, as the lowest one makes sure the higher
one(s) are never reached.
I looked more closely the DT and there is a misunderstanding of
the thermal framework in the definition.
There is one trip point with the passive type and the cpu cooling
device, followed by a second trip point with the active type *but*
the same cpu cooling device. That is wrong.
And finally, there is the hot trip point as a third mapping and
the same cooling device.
The hot trip point is only there to notify userspace and let it
take an immediate action to prevent an emergency shutdown when
reaching the critical temperature.
quoted
It can be applied only at 1 trip point. Multiple trip points is
only usefully for fan control to make sure the fan is not too
noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass
the lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
Yes, you are right. It should be something like (verbatim copy):
From: Eric Woudstra <hidden> Date: 2021-06-24 13:29:45
The SOC runs unthrotlled slowly to 80 degrees. This takes minutes. Polling interval 1 second or less does not matter much when looking at these temperature rise times
After that in more then an hour it slowly creeps up to 85. I believe the design is so that the SOC, under normal circumstances, can run at 1.35 GHz without throttling frequency, without heatsink. It just needs a safeguard for different circumstances.
Most of these SOCs can also run in industrial grade circumstances, which means up to 85 degrees ambient temperature already . If not industrial then this would be 60 degrees ambient already
But only someone at Mediatek can confirm this
Get BlueMail for Android
On Jun 24, 2021, 12:21 PM, at 12:21 PM, Daniel Lezcano [off-list ref] wrote:
Armada38x maximum die temperature 115 degrees Celcius. They really
get hotter then 100.
quoted
But for mt7622 I cannot find this value
Found that:
https://download.kamami.pl/p579344-MT7622A_Datasheet_for_BananaPi_Only%281%29.pdf
Chapter 3.3 - Thermal Characteristics
Given the values I suggest:
- Passive - 80°C
- Hot - 90°C
- Critical - 100°C
And passive polling set to 250ms.
It sounds like the sensor is not supporting the interrupt mode yet, so
a
big gap is needed with the Tj IMO to give the time to detect the trip
point crossing with the polling.
quoted
Get BlueMail for Android
On Jun 23, 2021, 10:08 PM, at 10:08 PM, Daniel Lezcano
[off-list ref] wrote:
quoted
quoted
On 23/06/2021 20:43, Eric Woudstra wrote:
quoted
I choose "hot" before, because 87 degrees seems ok to start
frequency
quoted
quoted
quoted
throttling. But, yes, it should be passive.
87 is still quite low if I compare this temperature with the
wrt3200acm Marvell dual core arm soc. They even went above 100
degrees so I feel for an arm processor inside a router box it is
fine
quoted
quoted
quoted
to use 87 degrees But maybe someone at Mediatek can give some more
details about operating temperatures.
Sometimes, the SoC vendor puts a high temperature in the DT just to
export the thermal zone and deal with it from userspace. So putting
the
quoted
quoted
high temp allow the userspace (usually a thermal engine - Android
stuff)
to deal with the mitigation without a kernel interaction.
Having more than 100°C could be this kind of setup. Only the
operating
quoted
quoted
temperature from the hardware documentation will tell the safe
temperature for the silicon.
IMO, 77°C is a good compromise until getting the documented temp.
87°C
quoted
quoted
sounds to me a bit too hot.
quoted
It may be possible to leave the active map in the device tree as
some
quoted
quoted
quoted
users of the bananapi might choose to install a fan as it is one of
the options.
The active trip only makes sense if the cooling device is a fan (or
any
quoted
quoted
active device), so the mapping points to a fan node, like:
If there is no such [pwm] fan output on the board, no active trip
point
quoted
quoted
should be added.
quoted
Get BlueMail for Android
On Jun 23, 2021, 5:58 PM, at 5:58 PM, Daniel Lezcano
[off-list ref] wrote:
quoted
On 23/06/2021 17:35, Eric Woudstra wrote:
quoted
It is only useful to set 1 map with the regulated temperature for
cpu frequency throttling. Same as in the kernel document
example.
It has no use to set frequency scaling on 2 different
temperature trip points, as the lowest one makes sure the higher
one(s) are never reached.
I looked more closely the DT and there is a misunderstanding of
the thermal framework in the definition.
There is one trip point with the passive type and the cpu cooling
device, followed by a second trip point with the active type *but*
the same cpu cooling device. That is wrong.
And finally, there is the hot trip point as a third mapping and
the same cooling device.
The hot trip point is only there to notify userspace and let it
take an immediate action to prevent an emergency shutdown when
reaching the critical temperature.
quoted
It can be applied only at 1 trip point. Multiple trip points is
only usefully for fan control to make sure the fan is not too
noisy when it is not necessary to be noisy.
The CPU will almost come to a dead stop when it starts to pass
the lowest thermal map with frequency throttling.
This is why it is a bug and needs a fix, not only adjustment.
Yes, you are right. It should be something like (verbatim copy):
From: Frank Wunderlich <hidden> Date: 2021-06-25 08:17:01
Hi,
Gesendet: Donnerstag, 24. Juni 2021 um 15:29 Uhr
Von: "Eric Woudstra" [off-list ref]
The SOC runs unthrotlled slowly to 80 degrees. This takes minutes. Polling interval 1 second or less does not matter much when looking at these temperature rise times
After that in more then an hour it slowly creeps up to 85. I believe the design is so that the SOC, under normal circumstances, can run at 1.35 GHz without throttling frequency, without heatsink. It just needs a safeguard for different circumstances.
Most of these SOCs can also run in industrial grade circumstances, which means up to 85 degrees ambient temperature already . If not industrial then this would be 60 degrees ambient already
But only someone at Mediatek can confirm this
maybe Matthias knows anybody?
get_maintainers-script shows no mtk employee for mtk_thermal driver, added Sean and Ryder as common Linux-Contacts...
Daniel from openwrt have some other mt7622 Boards maybe he can test the Fan approach below
On Jun 24, 2021, 12:21 PM, at 12:21 PM, Daniel Lezcano [off-list ref] wrote:
maybe adding FAN (r64, don't know for other mt7622 boards) for lower 2 trips (with adjusted temperature points) and cpu-throtteling for upper 2 trips
something like this (used the 70/80 trip points discussed before):
From: Daniel Golle <daniel@makrotopia.org> Date: 2021-06-25 09:40:18
On Fri, Jun 25, 2021 at 10:16:43AM +0200, Frank Wunderlich wrote:
Hi,
quoted
Gesendet: Donnerstag, 24. Juni 2021 um 15:29 Uhr
Von: "Eric Woudstra" [off-list ref]
The SOC runs unthrotlled slowly to 80 degrees. This takes minutes. Polling interval 1 second or less does not matter much when looking at these temperature rise times
After that in more then an hour it slowly creeps up to 85. I believe the design is so that the SOC, under normal circumstances, can run at 1.35 GHz without throttling frequency, without heatsink. It just needs a safeguard for different circumstances.
Most of these SOCs can also run in industrial grade circumstances, which means up to 85 degrees ambient temperature already . If not industrial then this would be 60 degrees ambient already
But only someone at Mediatek can confirm this
maybe Matthias knows anybody?
get_maintainers-script shows no mtk employee for mtk_thermal driver, added Sean and Ryder as common Linux-Contacts...
Daniel from openwrt have some other mt7622 Boards maybe he can test the Fan approach below
I got Linksys E8450 aka. Belkin RT3200 ( https://fcc.io/K7S-03571 ) as
well as Ubiquiti UniFi 6 LR ( https://fcc.io/SWX-U6LR ). Both got quite
massive customized heatsinks (see internal photos on FCC submission),
which results in much better heat dissipation than just having the
naked chip like on the BPi-R64.
Hence I also can't test the fan approach on boards other than the R64.
quoted hunk
quoted
On Jun 24, 2021, 12:21 PM, at 12:21 PM, Daniel Lezcano [off-list ref] wrote:
maybe adding FAN (r64, don't know for other mt7622 boards) for lower 2 trips (with adjusted temperature points) and cpu-throtteling for upper 2 trips
something like this (used the 70/80 trip points discussed before):
From: Frank Wunderlich <hidden> Date: 2021-06-25 09:32:18
Gesendet: Freitag, 25. Juni 2021 um 11:22 Uhr
Von: "Daniel Golle" [off-list ref]
On Fri, Jun 25, 2021 at 10:16:43AM +0200, Frank Wunderlich wrote:
quoted
Daniel from openwrt have some other mt7622 Boards maybe he can test the Fan approach below
I got Linksys E8450 aka. Belkin RT3200 ( https://fcc.io/K7S-03571 ) as
well as Ubiquiti UniFi 6 LR ( https://fcc.io/SWX-U6LR ). Both got quite
massive customized heatsinks (see internal photos on FCC submission),
which results in much better heat dissipation than just having the
naked chip like on the BPi-R64.
Hence I also can't test the fan approach on boards other than the R64.
Do your both mt7622 boards miss the fan-socket or is it not connected to pwm3? then we need to move the fan-parts to mt7622-bananapi-r64.dts instead of mt7622.dtsi
regards Frank
From: Daniel Golle <daniel@makrotopia.org> Date: 2021-06-25 10:11:46
Hi Frank,
On Fri, Jun 25, 2021 at 11:31:59AM +0200, Frank Wunderlich wrote:
quoted
Gesendet: Freitag, 25. Juni 2021 um 11:22 Uhr
Von: "Daniel Golle" [off-list ref]
quoted
On Fri, Jun 25, 2021 at 10:16:43AM +0200, Frank Wunderlich wrote:
quoted
Daniel from openwrt have some other mt7622 Boards maybe he can test the Fan approach below
I got Linksys E8450 aka. Belkin RT3200 ( https://fcc.io/K7S-03571 ) as
well as Ubiquiti UniFi 6 LR ( https://fcc.io/SWX-U6LR ). Both got quite
massive customized heatsinks (see internal photos on FCC submission),
which results in much better heat dissipation than just having the
naked chip like on the BPi-R64.
Hence I also can't test the fan approach on boards other than the R64.
Do your both mt7622 boards miss the fan-socket or is it not connected to pwm3? then we need to move the fan-parts to mt7622-bananapi-r64.dts instead of mt7622.dtsi
There is no fan intended on both devices.
E8450 has an unknown connector which **could** be for a fan, but I
never tried if and how it is actually connected to the SoC.
It could as well be an additional USB 2.0 (as it got 4 pins).
Hence I suggest to add the fan on PWM3 for the BPi-R64 only for now.
Cheers
Daniel
From: Daniel Lezcano <hidden> Date: 2021-06-25 09:57:10
Hi Frank,
On 25/06/2021 10:16, Frank Wunderlich wrote:
Hi,
quoted
Gesendet: Donnerstag, 24. Juni 2021 um 15:29 Uhr Von: "Eric
Woudstra" [off-list ref] The SOC runs unthrotlled slowly to
80 degrees. This takes minutes.
Polling interval 1 second or less does not matter much when looking at
these temperature rise times
quoted
After that in more then an hour it slowly creeps up to 85. I
believe
the design is so that the SOC, under normal circumstances, can run at
1.35 GHz without throttling frequency, without heatsink. It just needs a
safeguard for different circumstances.
quoted
Most of these SOCs can also run in industrial grade circumstances,
which means up to 85 degrees ambient temperature already . If not
industrial then this would be 60 degrees ambient already
quoted
But only someone at Mediatek can confirm this
maybe Matthias knows anybody? get_maintainers-script shows no mtk
employee for mtk_thermal driver,
added Sean and Ryder as common Linux-Contacts...
Daniel from openwrt have some other mt7622 Boards maybe he can test
the Fan approach below
quoted
On Jun 24, 2021, 12:21 PM, at 12:21 PM, Daniel Lezcano
Chapter 3.3 - Thermal Characteristics
Given the values I suggest:
- Passive - 80°C
- Hot - 90°C
- Critical - 100°C
maybe adding FAN (r64, don't know for other mt7622 boards) for lower
2
trips (with adjusted temperature points) and cpu-throtteling for upper 2
trips
It depends what you want to achieve first:
- better / sustained performance, then fan before
- quiet device or power saving (on battery) then cpu throttling before
That is board specific, it should be tuned on DT board specific file.
Some comments below:
quoted hunk
something like this (used the 70/80 trip points discussed before):
fan == active trip point
This is referring to the passive trip point. So it should point to the
CPU as it is now. Note the order of mitigation is inverted regarding the
proposal description.
fan == active trip point
This is referring to the passive trip point. So it should point to the
CPU as it is now. Note the order of mitigation is inverted regarding the
proposal description.
but we need to disable the passive trip as cpu-trotteling starts there...the higher temperature trips are currently not reached
summary
moving fan and cpu_thermal-override to bananapi-r64.dts
passive-trip: cooling-device = <&cpu0/1 0 0> as in erics Patch
active trip: cooling-device = <&fan0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
the other 2 unchanged
but i suggest changing the temperature points in mt7622 dtsi as this is SoC specific
so basicly:
From: Eric Woudstra <hidden> Date: 2021-06-25 11:07:48
I choose "hot* with CPU, because it was the best temperature. But it should really be passive only with the cooling device CPU but with a much higher temperature. For me 87 degrees is fine and tested. But for mainline we would better ask Mediatek for the correct maximum temperature.
Get BlueMail for Android
On Jun 25, 2021, 1:03 PM, at 1:03 PM, Frank Wunderlich [off-list ref] wrote:
quoted hunk
Hi
quoted
Gesendet: Freitag, 25. Juni 2021 um 11:57 Uhr
Von: "Daniel Lezcano" [off-list ref]
quoted
You should not add the fan in the mt7622.dtsi itself but in the board
specific file where there is a fan output on it. mt7622.dtsi is
fan == active trip point
This is referring to the passive trip point. So it should point to
the
quoted
CPU as it is now. Note the order of mitigation is inverted regarding
the
quoted
proposal description.
but we need to disable the passive trip as cpu-trotteling starts
there...the higher temperature trips are currently not reached
summary
moving fan and cpu_thermal-override to bananapi-r64.dts
passive-trip: cooling-device = <&cpu0/1 0 0> as in erics Patch
active trip: cooling-device = <&fan0 THERMAL_NO_LIMIT
THERMAL_NO_LIMIT>;
the other 2 unchanged
but i suggest changing the temperature points in mt7622 dtsi as this is
SoC specific
so basicly:
fan == active trip point
This is referring to the passive trip point. So it should point to the
CPU as it is now. Note the order of mitigation is inverted regarding the
proposal description.
but we need to disable the passive trip as cpu-trotteling starts there...the higher temperature trips are currently not reached
Sorry, can you rephrase it ? I'm not getting the point.
quoted hunk
summary
moving fan and cpu_thermal-override to bananapi-r64.dts
passive-trip: cooling-device = <&cpu0/1 0 0> as in erics Patch
active trip: cooling-device = <&fan0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
the other 2 unchanged
but i suggest changing the temperature points in mt7622 dtsi as this is SoC specific
so basicly:
hysteresis = <2000>;
type = "passive";
};
cpu_active: cpu-active {
- temperature = <67000>;
+ temperature = <80000>;
hysteresis = <2000>;
type = "active";
};
Move the active trip 'cpu_active' to mt7622-bananapi-bpi-r64.dts. and
set it to 70°C in the mt7622-bananapi-bpi-r64.dts, so the fan will act
before the cpu throttling.
The behavior should be the following: The temperature reaches 70°C, the
fan will start, if the temperature continues to increase, it will
increase the speed. If the temperature reaches 75°C, the fan is still
rotating at full speed but the cpu begins to be throttled.
AFAIU, it is a Cortex-A53 running @1.35GH, so except the board is in a
black metal box under the sun, I don't see how we can reach this thermal
limits.
From: Frank Wunderlich <hidden> Date: 2021-06-25 12:28:58
Gesendet: Freitag, 25. Juni 2021 um 13:47 Uhr
Von: "Daniel Lezcano" [off-list ref]
quoted
but we need to disable the passive trip as cpu-trotteling starts there...the higher temperature trips are currently not reached
Sorry, can you rephrase it ? I'm not getting the point.
the problem currently is that passive is at 47degress Celsius and trottles cpu, active (67°C) and hot points are never reached this way. so at least we need to change temperatures in dtsi, and maybe disable cpu-trotteling on passive trip. imho fan will never start if it is in active and cpu is trottled before in passive
quoted
summary
moving fan and cpu_thermal-override to bananapi-r64.dts
passive-trip: cooling-device = <&cpu0/1 0 0> as in erics Patch
active trip: cooling-device = <&fan0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
the other 2 unchanged
but i suggest changing the temperature points in mt7622 dtsi as this is SoC specific
so basicly:
hysteresis = <2000>;
type = "passive";
};
cpu_active: cpu-active {
- temperature = <67000>;
+ temperature = <80000>;
hysteresis = <2000>;
type = "active";
};
Move the active trip 'cpu_active' to mt7622-bananapi-bpi-r64.dts. and
set it to 70°C in the mt7622-bananapi-bpi-r64.dts, so the fan will act
before the cpu throttling.
The behavior should be the following: The temperature reaches 70°C, the
fan will start, if the temperature continues to increase, it will
increase the speed. If the temperature reaches 75°C, the fan is still
rotating at full speed but the cpu begins to be throttled.
passive to 75 and active lower to 70? is this as intended that active comes before passive?
mt7622-bananapi-bpi-r64.dts:
&cpu_thermal {
trips {
cpu_passive: cpu-passive {
temperature = <75000>;
hysteresis = <2000>;
type = "passive";
};
cpu_active: cpu-active {
temperature = <70000>;
hysteresis = <2000>;
type = "active";
};
};
cooling-maps {
map1 {
trip = <&cpu_active>;
cooling-device = <&fan0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
};
};
};
AFAIU, it is a Cortex-A53 running @1.35GH, so except the board is in a
black metal box under the sun, I don't see how we can reach this thermal
limits.
From: Daniel Lezcano <hidden> Date: 2021-06-25 12:50:48
On 25/06/2021 14:28, Frank Wunderlich wrote:
quoted
Gesendet: Freitag, 25. Juni 2021 um 13:47 Uhr
Von: "Daniel Lezcano" [off-list ref]
quoted
quoted
but we need to disable the passive trip as cpu-trotteling starts there...the higher temperature trips are currently not reached
Sorry, can you rephrase it ? I'm not getting the point.
the problem currently is that passive is at 47degress Celsius and
trottles cpu, active (67°C) and hot points are never reached this way.
so at least we need to change temperatures in dtsi, and maybe disable
cpu-trotteling on passive trip. imho fan will never start if it is in
active and cpu is trottled before in passive
Ok, thanks for the clarification.
quoted
quoted
summary
moving fan and cpu_thermal-override to bananapi-r64.dts
passive-trip: cooling-device = <&cpu0/1 0 0> as in erics Patch
active trip: cooling-device = <&fan0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
the other 2 unchanged
but i suggest changing the temperature points in mt7622 dtsi as this is SoC specific
so basicly:
hysteresis = <2000>;
type = "passive";
};
cpu_active: cpu-active {
- temperature = <67000>;
+ temperature = <80000>;
hysteresis = <2000>;
type = "active";
};
Move the active trip 'cpu_active' to mt7622-bananapi-bpi-r64.dts. and
set it to 70°C in the mt7622-bananapi-bpi-r64.dts, so the fan will act
before the cpu throttling.
The behavior should be the following: The temperature reaches 70°C, the
fan will start, if the temperature continues to increase, it will
increase the speed. If the temperature reaches 75°C, the fan is still
rotating at full speed but the cpu begins to be throttled.
passive to 75 and active lower to 70? is this as intended that active comes before passive?
Yes. So there is a default passive mitigation temp for the SoC at 75°C.
And the bpi has a setup with a fan mitigating before the cpu throttling.
mt7622-bananapi-bpi-r64.dts:
&cpu_thermal {
trips {
cpu_passive: cpu-passive {
temperature = <75000>;
hysteresis = <2000>;
type = "passive";
};
No need to add this trip point, it should be changed to 75°C in SoC DT
mt7622.dtsi. This fragment of DT will concatenate with the previous one.
AFAIU, it is a Cortex-A53 running @1.35GH, so except the board is in a
black metal box under the sun, I don't see how we can reach this thermal
limits.