Re: [PATCH v2] nvme: Add hardware monitoring support

[Guenter Roeck]

On 10/30/19 7:05 AM, Christoph Hellwig wrote:
> On Wed, Oct 30, 2019 at 08:16:48PM +0900, Akinobu Mita wrote:
> > The nvme_init_identify() can be called multiple time in nvme ctrl's
> > lifetime (e.g 'nvme reset /dev/nvme*' or suspend/resume paths), so
> > should we need to prevent nvme_hwmon_init() from registering hwmon
> > device more than twice?
> >
> > In the nvme thermal zone patchset[1], thernal zone is registered in
> > nvme_init_identify and unregistered in nvme_stop_ctrl().
>
> So Guenter said above the thermal subsystem could use the information
> from hwmon as well.  Does this mean this patch would solve your needs
> as well?
Depends on the requirements. Unlike hwmon/iio, we don't have clear
guidelines describing when thermal vs. hwmon would be a better choice.
There is some interconnect between thermal and hwmon, but quite often
it is a one-way street (hwmon devices can easily register thermal
zones, for thermal zone devices it is a bit more difficult to register
associated hwmon devices).

For the most part, peripherals (memory, network devices, video
controllers, real time clocks, etc) are today handled by the hardware
monitoring subsystem. The one notable exception is the ath10k wireless
controller, but even that registers both a thermal device and a hardware
monitoring device. Sometimes peripheral devices tell the hardware
monitoring subsystem that it should also register thermal zones (I
would guess that ath10k doesn't do that because the mechanism didn't
exist back in 2014). On the other side, SoCs typically register
thermal zones and rarely register as hardware monitoring device.

Guenter