phosphor-hwmon bottleneck potential

[Patrick Williams]

On Fri, May 05, 2017 at 11:50:52AM -0700, Robi Buranyi wrote:
> We could change the driver(s) to periodically poll the sensors like every
> second, and store the latest value. Then the sysfs read shall always return
> the last completed measurement data. We should do this for both the
> tachometers reading and the temperature sensors. There shall be some logic
> implemented for aging (when sensor read constantly fails)

This could certainly be done for the Aspeed PWM driver in question and
by my reading this might be the best approach.  Due to the way that
hardware works it seems like you'd an interval between samples of at
least 1 second in order to be accurate to roughly 100 RPMs.  I don't
know what accuracy you are striving for, but blocking an application on
a syscall for 1 second doesn't sound like a great implementation.

We could check with Guenter Roeck, the HWMON maintainer, to see if he
has any recommendations on the best way to handle this driver.

Changing HWMON on the whole to do this (the '(s)' part of your 'driver(s)'
statement) seems pretty out of scope and not likely to be accepted by
upstream.  Most HWMON drivers do simple IO in the read syscall, like a
few i2c operations to read a register out in a device so there isn't
problem with a long occurring block.

-- 
Patrick Williams
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: Digital signature
URL: <http://lists.ozlabs.org/pipermail/openbmc/attachments/20170505/fb4c442d/attachment.sig>