12:31 a.m.
On Fri, Feb 26, 2016 at 11:44 PM, Pandruvada, Srinivas <
srinivas.pandruvada(a)intel.com> wrote:
On Fri, 2016-02-26 at 11:41 +0530, Rajagopal Aravindan wrote:
> Hello,
>
> I am looking at using thermald for an embedded platform.
> Can someone please point me to where I can get answers for the below
> ...
> How to get the list of devices that are currently managed by thermald
> (including those yet to be throttled), given that NO ACPI style xml
> is used. ?
If this is not a PC style device, then you need to create a config
file.
man thermal-conf.xml to see detail.
Basically you need to understand:
- What all the devices you want to manage for thermal conditions? Do
you have a temperature sensor on that device or close the device? If
yes how will you read temperature? Is there sysfs path to read the
temperature?
- What temperature you want to take action?
- What are the operation you can do to influence the device to reduce
temperature? Reduce performance or activate some device like a FAN or
increase speed of a FAN. Do you have sysf entry to do this.
For CPUs temperature it is little simple as you can use cpufreq to
reduce performance, which is supported for most of the embedded procees
sors also.
Thanks,
Srinivas
> How to go about adding new devices to the list of currently managed
> ones. Where to start from ?
>
> Thanks,
> Rajagopal
> _______________________________________________
> Linux-Thermal-Daemon mailing list
> Linux-Thermal-Daemon(a)lists.01.org
> https://lists.01.org/mailman/listinfo/linux-thermal-daemon
Srinivas,
Following your above heads-up on this, I managed to gather the below
information regarding my task.
1. A SoC component needs to be thermally controlled.
2. Its sensor is exposed via sysfs, which is async capable, through
which its temperature can be read.
3. The temp threshold values have been provided to me.
4. I would be reducing that device's performance (passive), to bring
down its temperature.
5. There is NO sysfs entry for reducing the device's performance, it
will be using a proprietary method.
Given the above details, could you please confirm my below understanding ?
1. Temp sensing and triggering of action on threshold breach, will be
taken care of by existing thermald framework (sysfs).
2. For initiating the cooling action (passive) ..
1. I would have to write my own cooling device class, which extends
the thermald's cooling device class.
2. Create an object for it and add it to the global list of cooling
devices.
3. Implement the proprietary throttling logic in its *set_curr_state*
member function.
4. Put a *CoolingDevice* entry corresponding to the *SensorType*, as
found in sysfs.
5. The *type* inside *CoolingDevice* should match with what was
passed to *set_cdev_type*.
If there is a way to accomplish this without having to write a new class,
it would be helpful if you could please provide some pointers to that.
Thanks & Regards,
Rajagopal
8:50 a.m.
New subject: Extending thermald to custom devices on embedded platform
On 03/14/2016 06:31 AM, Rajagopal Aravindan wrote:
On Fri, Feb 26, 2016 at 11:44 PM, Pandruvada, Srinivas
<srinivas.pandruvada(a)intel.com <mailto:srinivas.pandruvada@intel.com>>
wrote:
On Fri, 2016-02-26 at 11:41 +0530, Rajagopal Aravindan wrote:
> Hello,
>
> I am looking at using thermald for an embedded platform.
> Can someone please point me to where I can get answers for the below
> ...
> How to get the list of devices that are currently managed by
thermald
> (including those yet to be throttled), given that NO ACPI style xml
> is used. ?
If this is not a PC style device, then you need to create a config
file.
man thermal-conf.xml to see detail.
Basically you need to understand:
- What all the devices you want to manage for thermal conditions? Do
you have a temperature sensor on that device or close the device? If
yes how will you read temperature? Is there sysfs path to read the
temperature?
- What temperature you want to take action?
- What are the operation you can do to influence the device to reduce
temperature? Reduce performance or activate some device like a FAN or
increase speed of a FAN. Do you have sysf entry to do this.
For CPUs temperature it is little simple as you can use cpufreq to
reduce performance, which is supported for most of the embedded
procees
sors also.
Thanks,
Srinivas
> How to go about adding new devices to the list of currently managed
> ones. Where to start from ?
>
> Thanks,
> Rajagopal
> _______________________________________________
> Linux-Thermal-Daemon mailing list
> Linux-Thermal-Daemon(a)lists.01.org
<mailto:Linux-Thermal-Daemon@lists.01.org>
> https://lists.01.org/mailman/listinfo/linux-thermal-daemon
Srinivas,
Following your above heads-up on this, I managed to gather the below
information regarding my task.
1. A SoC component needs to be thermally controlled.
2. Its sensor is exposed via sysfs, which is async capable, through
which its temperature can be read.
3. The temp threshold values have been provided to me.
4. I would be reducing that device's performance (passive), to bring
down its temperature.
5. There is NO sysfs entry for reducing the device's performance, it
will be using a proprietary method.
Given the above details, could you please confirm my below understanding ?
1. Temp sensing and triggering of action on threshold breach, will be
taken care of by existing thermald framework (sysfs).
2. For initiating the cooling action (passive) ..
1. I would have to write my own cooling device class, which
extends the thermald's cooling device class.
2. Create an object for it and add it to the global list of
cooling devices.
3. Implement the proprietary throttling logic in its
*/set_curr_state/* member function.
4. Put a /*CoolingDevice*/ entry corresponding to the
/*SensorType*/, as found in sysfs.
5. The /*type*/ inside /*CoolingDevice*/ should match with what
was passed to /*set_cdev_type*/.
If there is a way to accomplish this without having to write a new
class, it would be helpful if you could please provide some pointers
to that.
Since there is no sysfs entry for your cooling control, you have to
write a new class. I think you will not open source your cooling methods
as they are proprietary.
Ideally I would prefer to enhance the XML configuration to send a dbus
or binder message to remote task in the same way as current sysfs path
in XML thermal config.
Thanks,
Srinivas
Thanks & Regards,
Rajagopal
_______________________________________________
Linux-Thermal-Daemon mailing list
Linux-Thermal-Daemon(a)lists.01.org
https://lists.01.org/mailman/listinfo/linux-thermal-daemon
5:27 a.m.
New subject: Extending thermald to custom devices on embedded platform
On Tue, Mar 15, 2016 at 3:20 AM, Srinivas Pandruvada <
srinivas.pandruvada(a)linux.intel.com> wrote:
On 03/14/2016 06:31 AM, Rajagopal Aravindan wrote:
On Fri, Feb 26, 2016 at 11:44 PM, Pandruvada, Srinivas <
<srinivas.pandruvada@intel.com>srinivas.pandruvada@intel.com> wrote:
> On Fri, 2016-02-26 at 11:41 +0530, Rajagopal Aravindan wrote:
> > Hello,
> >
> > I am looking at using thermald for an embedded platform.
> > Can someone please point me to where I can get answers for the below
> > ...
> > How to get the list of devices that are currently managed by thermald
> > (including those yet to be throttled), given that NO ACPI style xml
> > is used. ?
> If this is not a PC style device, then you need to create a config
> file.
> man thermal-conf.xml to see detail.
>
> Basically you need to understand:
> - What all the devices you want to manage for thermal conditions? Do
> you have a temperature sensor on that device or close the device? If
> yes how will you read temperature? Is there sysfs path to read the
> temperature?
> - What temperature you want to take action?
> - What are the operation you can do to influence the device to reduce
> temperature? Reduce performance or activate some device like a FAN or
> increase speed of a FAN. Do you have sysf entry to do this.
> For CPUs temperature it is little simple as you can use cpufreq to
> reduce performance, which is supported for most of the embedded procees
> sors also.
>
> Thanks,
> Srinivas
>
> > How to go about adding new devices to the list of currently managed
> > ones. Where to start from ?
> >
> > Thanks,
> > Rajagopal
> > _______________________________________________
> > Linux-Thermal-Daemon mailing list
> > Linux-Thermal-Daemon(a)lists.01.org
> > https://lists.01.org/mailman/listinfo/linux-thermal-daemon
Srinivas,
Following your above heads-up on this, I managed to gather the below
information regarding my task.
1. A SoC component needs to be thermally controlled.
2. Its sensor is exposed via sysfs, which is async capable, through
which its temperature can be read.
3. The temp threshold values have been provided to me.
4. I would be reducing that device's performance (passive), to bring
down its temperature.
5. There is NO sysfs entry for reducing the device's performance, it
will be using a proprietary method.
Given the above details, could you please confirm my below understanding ?
1. Temp sensing and triggering of action on threshold breach, will be
taken care of by existing thermald framework (sysfs).
2. For initiating the cooling action (passive) ..
1. I would have to write my own cooling device class, which extends
the thermald's cooling device class.
2. Create an object for it and add it to the global list of cooling
devices.
3. Implement the proprietary throttling logic in its
*set_curr_state* member function.
4. Put a *CoolingDevice* entry corresponding to the *SensorType*,
as found in sysfs.
5. The *type* inside *CoolingDevice* should match with what was
passed to *set_cdev_type*.
If there is a way to accomplish this without having to write a new class,
it would be helpful if you could please provide some pointers to that.
Since there is no sysfs entry for your cooling control, you have to write
a new class. I think you will not open source your cooling methods as they
are proprietary.
Ideally I would prefer to enhance the XML configuration to send a dbus or
binder message to remote task in the same way as current sysfs path in XML
thermal config.
Thanks,
Srinivas
Thanks & Regards,
Rajagopal
<Linux-Thermal-Daemon(a)lists.01.org>
<https://lists.01.org/mailman/listinfo/linux-thermal-daemon>
_______________________________________________
Linux-Thermal-Daemon mailing list
Linux-Thermal-Daemon(a)lists.01.org
https://lists.01.org/mailman/listinfo/linux-thermal-daemon
Srinivas,
In all the sample XMLs I have come across so far, either sysfs path or a
custom device file path is specified.
Am I to understand there are also XML tags to specify d-bus object path and
have a method under it invoked ?
Whether cooling is initiated actively or passively is a cooling device
design choice.
Why is that information also to be provided for <TripPoint> ?
Also, it appears that trip point violation doesn't always lead to cooling,
some times it activates frequent temp sensing. This was my run-time
observation on my target h/w.
It would be very helpful if you could point to some links / documents or
provide more details on trip points.
Thanks & Regards,
Rajagopal
7:58 a.m.
New subject: Extending thermald to custom devices on embedded platform
Srinivas,
In all the sample XMLs I have come across so far, either sysfs path
or a custom device file path is specified.
Am I to understand there are also XML tags to specify d-bus object
path and have a method under it invoked ?
This is not supported. If you want to add
support you can. It will be
useful.
Whether cooling is initiated actively or passively is a cooling
device design choice.
Why is that information also to be provided for <TripPoint> ?
<Preference>QUIET</Preference>
then only passive trip points will be active
<Preference>PERFORMANCE</Preference>
only active trip points will be active
The trip points (temperature where some action will take place) can be
different when you want more performance (you may activate a Fan) or
when want a quite mode using simply passive cooling.
Also, it appears that trip point violation doesn't always lead
to
cooling, some times it activates frequent temp sensing.
Internally based on the you
active/passive trip point internally we set
another trip point to activate polling. Sometimes when system is busy,
you will miss the uevents when you are running thermald without polling
for temperature. If your are polling for temperature change or you have
sensors which can't notify temperature change asynchronously then this
trip is not much value.
You can read ACPI 6.0 spec thermal handling chapter. The trip and zones
are modeled as defined here.
Thanks,
Srinivas
This was my run-time observation on my target h/w.
It would be very helpful if you could point to some links / documents
or provide more details on trip points.
Thanks & Regards,
Rajagopal
1:11 a.m.
New subject: Extending thermald to custom devices on embedded platform
On Wed, Mar 16, 2016 at 2:28 AM, Srinivas Pandruvada <
srinivas.pandruvada(a)linux.intel.com> wrote:
> Srinivas,
>
> In all the sample XMLs I have come across so far, either sysfs path
> or a custom device file path is specified.
>
> Am I to understand there are also XML tags to specify d-bus object
> path and have a method under it invoked ?
This is not supported. If you want to add support you can. It will be
useful.
>
> Whether cooling is initiated actively or passively is a cooling
> device design choice.
>
> Why is that information also to be provided for <TripPoint> ?
>
<Preference>QUIET</Preference>
then only passive trip points will be active
<Preference>PERFORMANCE</Preference>
only active trip points will be active
The trip points (temperature where some action will take place) can be
different when you want more performance (you may activate a Fan) or
when want a quite mode using simply passive cooling.
> Also, it appears that trip point violation doesn't always lead to
> cooling, some times it activates frequent temp sensing.
Internally based on the you active/passive trip point internally we set
another trip point to activate polling. Sometimes when system is busy,
you will miss the uevents when you are running thermald without polling
for temperature. If your are polling for temperature change or you have
sensors which can't notify temperature change asynchronously then this
trip is not much value.
You can read ACPI 6.0 spec thermal handling chapter. The trip and zones
are modeled as defined here.
Thanks,
Srinivas
> This was my run-time observation on my target h/w.
> It would be very helpful if you could point to some links / documents
> or provide more details on trip points.
>
> Thanks & Regards,
> Rajagopal
Srinivas,
Could you please provide your inputs on the below ?
1. *max *- I couldn't get info on this trip point type from ACPI 6.0
spec.
2. Is there any limit on the max no of trip points for an async-capable
sensor ?
3. Is there anything like a pure async mode ?
4. How to go about adding D-Bus support, for both sensing & throttling ?
1. Are there XML tags already available, for which parsing &
operational logic needs to be implemented ?
2. If we were to define new tags, will it have to conform to or
comply with any spec ?
3. Are there devices already available in market, which expose
sensing & throttling interfaces via D-Bus ?
Thanks & Regards,
Rajagopal
10:21 p.m.
New subject: Extending thermald to custom devices on embedded platform
On Thu, Mar 31, 2016 at 7:41 PM, Rajagopal Aravindan <
a.rajagopal.81(a)gmail.com> wrote:
On Wed, Mar 16, 2016 at 2:28 AM, Srinivas Pandruvada <
srinivas.pandruvada(a)linux.intel.com> wrote:
>
> > Srinivas,
> >
> > In all the sample XMLs I have come across so far, either sysfs path
> > or a custom device file path is specified.
> >
> > Am I to understand there are also XML tags to specify d-bus object
> > path and have a method under it invoked ?
> This is not supported. If you want to add support you can. It will be
> useful.
>
> >
> > Whether cooling is initiated actively or passively is a cooling
> > device design choice.
> >
> > Why is that information also to be provided for <TripPoint> ?
> >
> <Preference>QUIET</Preference>
> then only passive trip points will be active
> <Preference>PERFORMANCE</Preference>
> only active trip points will be active
>
> The trip points (temperature where some action will take place) can be
> different when you want more performance (you may activate a Fan) or
> when want a quite mode using simply passive cooling.
>
>
> > Also, it appears that trip point violation doesn't always lead to
> > cooling, some times it activates frequent temp sensing.
> Internally based on the you active/passive trip point internally we set
> another trip point to activate polling. Sometimes when system is busy,
> you will miss the uevents when you are running thermald without polling
> for temperature. If your are polling for temperature change or you have
> sensors which can't notify temperature change asynchronously then this
> trip is not much value.
>
> You can read ACPI 6.0 spec thermal handling chapter. The trip and zones
> are modeled as defined here.
>
> Thanks,
> Srinivas
>
> > This was my run-time observation on my target h/w.
> > It would be very helpful if you could point to some links / documents
> > or provide more details on trip points.
> >
> > Thanks & Regards,
> > Rajagopal
>
Srinivas,
Could you please provide your inputs on the below ?
1. *max *- I couldn't get info on this trip point type from ACPI 6.0
spec.
2. Is there any limit on the max no of trip points for an
async-capable sensor ?
3. Is there anything like a pure async mode ?
4. How to go about adding D-Bus support, for both sensing & throttling
?
1. Are there XML tags already available, for which parsing &
operational logic needs to be implemented ?
2. If we were to define new tags, will it have to conform to or
comply with any spec ?
3. Are there devices already available in market, which expose
sensing & throttling interfaces via D-Bus ?
Thanks & Regards,
Rajagopal
Srinivas,
Awaiting your inputs on this.
Thanks & Regards,
Rajagopal
2:18 a.m.
New subject: Extending thermald to custom devices on embedded platform
On Tue, 2016-04-12 at 17:51 +0530, Rajagopal Aravindan wrote:
On Thu, Mar 31, 2016 at 7:41 PM, Rajagopal Aravindan
<a.rajagopal.81@
gmail.com> wrote:
> On Wed, Mar 16, 2016 at 2:28 AM, Srinivas Pandruvada <srinivas.pand
> ruvada(a)linux.intel.com> wrote:
> > > Srinivas,
> > >
> > > In all the sample XMLs I have come across so far, either sysfs
> > path
> > > or a custom device file path is specified.
> > >
> > > Am I to understand there are also XML tags to specify d-bus
> > object
> > > path and have a method under it invoked ?
> > This is not supported. If you want to add support you can. It
> > will be
> > useful.
> >
> > >
> > > Whether cooling is initiated actively or passively is a cooling
> > > device design choice.
> > >
> > > Why is that information also to be provided for <TripPoint> ?
> > >
> > <Preference>QUIET</Preference>
> > then only passive trip points will be active
> > <Preference>PERFORMANCE</Preference>
> > only active trip points will be active
> >
> > The trip points (temperature where some action will take place)
> > can be
> > different when you want more performance (you may activate a Fan)
> > or
> > when want a quite mode using simply passive cooling.
> >
> >
> > > Also, it appears that trip point violation doesn't always lead
> > to
> > > cooling, some times it activates frequent temp sensing.
> > Internally based on the you active/passive trip point internally
> > we set
> > another trip point to activate polling. Sometimes when system is
> > busy,
> > you will miss the uevents when you are running thermald without
> > polling
> > for temperature. If your are polling for temperature change or
> > you have
> > sensors which can't notify temperature change asynchronously then
> > this
> > trip is not much value.
> >
> > You can read ACPI 6.0 spec thermal handling chapter. The trip and
> > zones
> > are modeled as defined here.
> >
> > Thanks,
> > Srinivas
> >
> > > This was my run-time observation on my target h/w.
> > > It would be very helpful if you could point to some links /
> > documents
> > > or provide more details on trip points.
> > >
> > > Thanks & Regards,
> > > Rajagopal
> >
> Srinivas,
>
Hi,
> Could you please provide your inputs on the below ?
> max - I couldn't get info on this trip point type from ACPI 6.0
> spec.
This is not defined in ACPI spec.
> Is there any limit on the max no of trip points for an async-
> capable sensor ?
No. The lowest trip will trigger polling.
> Is there anything like a pure async mode ?
No. As the
uevents reliability reduces depending on other load in the
system. So we will start polling.
> How to go about adding D-Bus support, for both sensing &
throttling
> ?
We have dbus Interface to control many thd_dbus_interface.xml.
These are request to thermald not from thermald. We don't have a way
send a dbus message from thermald to other process for throttling.
So you have to implement.
> Are there XML tags already available, for which parsing &
> operational logic needs to be implemented ?
> If we were to define new tags, will it have to conform to or comply
> with any spec ?
There is no spec.
> Are there devices already available in market, which expose
sensing
> & throttling interfaces via D-Bus ?
I am not aware of any.
Thanks,
Srinivas
>
> Thanks & Regards,
> Rajagopal
>
Srinivas,
Awaiting your inputs on this.
Thanks & Regards,
Rajagopal
2350
days inactive
2379
days old
linux-thermal-daemon@lists.01.org
6 comments
2 participants
participants (2)
-
Rajagopal Aravindan -
Srinivas Pandruvada