New Driver for electrical energy measurement

Fri Jul 14 09:40:39 UTC 2017

On 07/14/2017 11:00 AM, Guenter Roeck wrote:
> On 07/14/2017 01:07 AM, Jonathan Cameron wrote:
>> On Wed, 12 Jul 2017 15:19:40 +0200
>> Christian Gromm <christian.gromm at microchip.com> wrote:
>>
>>> On Wed, 12 Jul 2017 14:51:01 +0200
>>> Greg KH <gregkh at linuxfoundation.org> wrote:
>>>
>>>> On Wed, Jul 12, 2017 at 02:18:54PM +0200, Christian Gromm wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> Microchip is planning to introduce a driver for a new companion
>>>>> chip series capable of electical energy measurement. However, we
>>>>> are not sure which location in the source tree is the most
>>>>> suitable. First thought was to put it into /drivers/powercap. But
>>>>> now we are in discussions whether it would make even more sense to
>>>>> introduce a new driver for electrical energy measuring or
>>>>> monitoring.
>>>>
>>>> Why not just have it be an IIO device, I think they have energy
>>>> sensors already (adding the iio mailing list to find out...)
>>>
>>> We thought about hwmon, but not IIO.
>>>
>>>>   
>>>>>
>>>>> Following is a rough overview of the part we want the driver for.
>>>>> Any input on this will be appreciated.
>>>>>
>>>>> thanks,
>>>>> Chris
>>>>>
>>>>> === Introduction
>>>>>
>>>>> Following the recent focus on low-power systems, Microchip has
>>>>> designed a companion chip series capable of measuring the
>>>>> electrical energy flow of an electrical system.  The series in
>>>>> question, EM Chip  is capable of measuring the energy flow (in or
>>>>> out) of an electrical system by monitoring one or more power
>>>>> rails.  The 1st chip of the series, EM Chip is able to measure the
>>>>> net energy flow over 4 different power rails.  The chip itself is a
>>>>> small QFN16 (4mm x 4mm x 0.5mm) or WLCSP16 (2.25mm x 2.2mm) package.
>>>>>
>>>>> === Theory of Operation
>>>>>
>>>>> In order to measure the amount of energy entering/exiting a system,
>>>>> a small shunt resistor is interleaved on a power rail. By measuring
>>>>> the small voltage drop across the shunt resistor (a known value),
>>>>> the electrical current is measured.
>>>>>
>>>>>     I = U/R
>>>>>
>>>>> Depending on whether the system is consuming or producing energy,
>>>>> the measured current value can be either a positive or negative
>>>>> number.
>>>>>
>>>>> To get the instantaneous power figure we will have to measure the
>>>>> power rail’s voltage and multiply it with the measured value for
>>>>> the current
>>>>>
>>>>>     P = V * I = V * U_shunt/R
>>>>>
>>>>> Knowing the instantaneous power and by making a high enough sample
>>>>> rate (for measuring the current and the power rail voltage), we can
>>>>> assume the measured value for power is equal to the average power
>>>>> figure for the amount of time between 2 sampling moments.
>>>>>
>>>>> Now that we also know the average power for a given time interval
>>>>> (dt_x = time between 2 sampling moments; sampling speed is known),
>>>>> we can measure the energy amount entering or exiting the system
>>>>> between 2 sampling moments
>>>>>
>>>>>     E_partial_x = P * dt_x
>>>>>
>>>>> Having the energy information available, we can continue to add the
>>>>> subsequent energy values for as long as the system is active. The
>>>>> amount of energy is the sum of all the partial energy values
>>>>> measured for each time interval E = Sum (E_partial_x), where x can
>>>>> take values from 0 till infinite.
>>>>>
>>>>>
>>>>> === Chip Operation
>>>>>
>>>>> The EM Chip chip uses the same principles of operation as presented
>>>>> in the “Theory of Operation” and it does so for a number of 4
>>>>> channels. 4 independent power rails can be energy monitored in the
>>>>> same time.  The chip is controlled over I2C/SMBus by an I2C/SMBus
>>>>> master.
>>>>>
>>>>> In order to reduce the amount of information traffic between the EM
>>>>> Chip and the entity asking for its information, the chip features
>>>>> large accumulator energy registers for the accumulated energy
>>>>> values. It can accumulate up to 2^24 power values. Depending on the
>>>>> selected chip’s sampling speed, such a number of samples are
>>>>> produced in about 4.5 hours (fastest sampling speed) up to 582.5
>>>>> hours or more than 24 days (lowest sampling speed).
>>>>>
>>>>> EM Chip measures the power as a 28-bits number. The 28-bits number
>>>>> is the multiplication result of the 16-bits number used to measure
>>>>> the power rail voltage and the 12-bits number used for measuring
>>>>> the voltage drop across the power rail shunt resistor.
>>>>>
>>>>> If bidirectional energy flow is required, the power is measured as a
>>>>> 27-bits number and 1 bit for the sign. When no bidirectional flow is
>>>>> needed, the power value is measured as an unsigned 28-bits number.
>>>>>
>>>>> Due to the relatively large size of the accumulated energy
>>>>> registers inside the chip (48 bits), in the worst case scenario
>>>>> (power values are full scale numbers), a number of 2^20 full-scale
>>>>> power values can be measured before energy register’s overflow.
>>>>> Using the fastest sampling speed, the accumulated energy registers
>>>>> overflows only after a bit over 17 minutes, while at the lowest
>>>>> sampling speed, it would overflow in over 36 hours.
>>>>>
>>>>> Thus, the chip requires infrequent reads of the accumulated energy
>>>>> registers. Even in the worst case scenario, the time between 2
>>>>> consecutive energy accumulator reads can be over 17 minutes !
>>>>>
>>>>> In order to keep a longer history of energy measurements, an
>>>>> I2C/SMBus master (e.g. SoC) would read the accumulated energy
>>>>> register values and then use larger “soft” accumulated energy
>>>>> registers to extend the maximum overflow period.
>>>>>
>>>>> The EM Chip chip can monitor rails up to 32V. It can monitor the
>>>>> energy amounts used by various sub-components of a system (e.g.
>>>>> CPU, GPU, memory, hard-drives, USB ports, backlight, wireless
>>>>> adapters, cameras, displays, …)
>>>>>
>>>>> EM Chip is able to start operating immediately after power-up with
>>>>> no CPU intervention at all. Such a feature is useful, because it
>>>>> can show the amount of energy consumed by a system before the
>>>>> latter finished booting its operating system.
>>>>>
>>>>> The chip’s own current draw is around 20uA (in low-power mode,
>>>>> lowest sampling speed - 8 samples/sec) to 675uA (for the highest
>>>>> sampling speed - 1024 samples/sec)
>>>>>
>>>>> When operated in low-power mode, it can be used to monitor the
>>>>> stand-by energy draw of the system. As an example, such a mode is
>>>>> useful when a system is suspended to RAM or to measure the energy
>>>>> usage from the power on till the booting process is finished.
>>>>>
>>>>> === Linux Driver
>>>>>
>>>>> While the chip is due to be publicly released in Q3 2017, a
>>>>> selected number of PC OEM manufacturers will include one or more EM
>>>>> Chip chips on their systems.  We would like to include a driver for
>>>>> EM Chip chip and its follow-up products, such that Linux Kernel
>>>>> will be able to provide the energy information as soon as computing
>>>>> systems using this series of chips will become available.
>>>>>
>>>>> By providing accurate energy measurements, the computing systems
>>>>> along with their operating systems will be able to run more
>>>>> efficiently.
>>>>
>>>>
>>>> Do you have any kernel code already to show how this will get hooked
>>>> up to the device?  Do you need device tree bindings for the sensors,
>>>> or are they on a discoverable bus?
>>>
>>> No, I don't. I wanted to bring this up on the mailing list first,
>>> before I get started.
>>> Sensors are attached via I2C.
>>
>> Sure, I don't blame you having myself gone through several subsystems with
>> a driver in the past!
>>
>> As Greg mentioned there are existing IIO drivers doing similar power
>> measurements.  So yes, a driver for such a part would be welcome.
>>
>> I can see we may be 'straining' some of the interfaces a bit for those
>> 48 bit values, but we can probably do something simple using the
>> two 32 bit values available from a raw read in IIO.
>>
>> Going to get awkward the first time we get a 65+ bit device though.
>> There goes my assumptions again :(
>>
>> Anyhow, nice sounding part so looking forward to seeing some code when
>> you are ready.
>>
>> Note we do have additional energy drivers still in staging (dating back
>> to the early days of IIO) but those are 3 phase units so not so relevant
>> (and their interfaces need some work which is why they are still in
>> staging).
>>
>> The adc/ina2xx_adc.c driver is probably a better place to start.
>>
>> Hmm. The only thing that makes me doubt IIO as the obvious place is
>> that the focus is clearly on PC energy monitoring.  Now you can bridge
>> IIO to hwmon but I'd like a bit of input from Guenter on
>> whether that is a sensible approach here.
>>
> 
> I don't think I have enough information for a sensible answer. What
> else besides raw energy measurement do the chips support ? Do they have
> limits / alarms ? Do they measure voltage and current as well ?
> If not, iio would be the better place. If the chips support PMBus
> (which would make sense, but it doesn't sound like it), hwmon would
> obviously be better, since it already has the necessary infrastructure.

There is also some precedence for these kinds of devices in the power-supply
framework. E.g. the LTC2941 and similar work in a similar way to whats
described above
(https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/power/supply/ltc2941-battery-gauge.c
http://cds.linear.com/docs/en/datasheet/2941fb.pdf). Isn't it lovely to have
so much choice?