On 8/2/23 23:34, Nilawar, Badal wrote:
There should be just one hardware monitoring device. Just use energyN
and reference the input with an appropriate sensor label.
Idea was to expose energy attributes under saperate hwmon folder with device names xe_gtN. But with channel/label approach it will look like energyN_input (energy1_input - device, energy_2,3input - gt0/gt1) with appropriate energyN_label (energy1_label = "energy device", energy_2,3 = "energy gt0/gt1". With this approach we can avoid using 2 structures xe_hwmon and xe_hwmon_data.
There is really no such thing as "energy device". I'd suggest to find a better name,
such as "Package energy" or just "Package".
Anyway, your code is flawed: It claims to handle energy overflows, but it doesn't
really do that reliably. The returned numbers are only reliable if read at least
once within an overflow interval. Making it reliable would require introducing
a worker or kernel thread which reads the data from the chip in regular intervals
and accumulates independently of attribute reads.
Guenter
