Where: org.metricshub.ipmi.core.coding.commands.sdr (new command), src/main/java/org/metricshub/ipmi/client/runner/GetSensorsRunner.java, src/main/java/org/metricshub/ipmi/core/coding/commands/sdr/record/FullSensorRecord.java.
What happens: For a Full Sensor record whose linearization is non-linear (byte 24 = 70h-7Fh, IPMI 2.0 Table 43-1), the M, B and exponent factors of the SDR hold at the nominal reading only. Converting an arbitrary raw reading needs the Get Sensor Reading Factors command (section 35.5, NetFn Sensor/Event, command 23h), which returns the factors valid for a given raw reading. The library does not implement it.
Since #143, such sensors fail safe: FullSensorRecord.hasAnalogReading() is false, calcFormula() returns NaN, and IpmiClient returns the sensor with its states but without reading (1.2.02 dropped the sensor altogether). The gap noted in #83 is therefore closed by not reporting a value, not by computing it.
Suggested implementation:
- A
GetSensorReadingFactors coder: request = sensor number, raw reading byte; response (Table 35-6) = next reading, M (10 bits, 2's complement), tolerance (6 bits), B (10 bits, 2's complement), accuracy (10 bits), accuracy exponent, R exponent, B exponent, in the same layout as bytes 25-30 of the Full Sensor record.
- A conversion entry point on
FullSensorRecord that takes the factors as arguments (the current calcFormula() uses the record's own), applying the record's linearization function (e^x, log, ...) on top of the returned linear factors.
GetSensorsRunner: for a Full record whose linearization is 70h-7Fh, call Get Sensor Reading Factors with the raw reading obtained from Get Sensor Reading, and convert with the returned factors. The thresholds of such a record would need the same treatment (one call per threshold byte) if they are to be reported.hasAnalogReading() then stays false only for the data format 11b.
Evidence: by the specification. Neither test BMC (GIGABYTE, Lenovo IMM) has a non-linear sensor, so the implementation can only be verified with unit tests on crafted bytes; say so in the PR.
Related: #83 (closed by #143 with the fail-safe behaviour), #111 and #101 (other client API enhancements). Low priority: non-linear sensors are rare in practice.
Where:
org.metricshub.ipmi.core.coding.commands.sdr(new command),src/main/java/org/metricshub/ipmi/client/runner/GetSensorsRunner.java,src/main/java/org/metricshub/ipmi/core/coding/commands/sdr/record/FullSensorRecord.java.What happens: For a Full Sensor record whose linearization is non-linear (byte 24 =
70h-7Fh, IPMI 2.0 Table 43-1), the M, B and exponent factors of the SDR hold at the nominal reading only. Converting an arbitrary raw reading needs the Get Sensor Reading Factors command (section 35.5, NetFn Sensor/Event, command23h), which returns the factors valid for a given raw reading. The library does not implement it.Since #143, such sensors fail safe:
FullSensorRecord.hasAnalogReading()isfalse,calcFormula()returnsNaN, andIpmiClientreturns the sensor with its states but without reading (1.2.02 dropped the sensor altogether). The gap noted in #83 is therefore closed by not reporting a value, not by computing it.Suggested implementation:
GetSensorReadingFactorscoder: request = sensor number, raw reading byte; response (Table 35-6) = next reading, M (10 bits, 2's complement), tolerance (6 bits), B (10 bits, 2's complement), accuracy (10 bits), accuracy exponent, R exponent, B exponent, in the same layout as bytes 25-30 of the Full Sensor record.FullSensorRecordthat takes the factors as arguments (the currentcalcFormula()uses the record's own), applying the record's linearization function (e^x,log, ...) on top of the returned linear factors.GetSensorsRunner: for a Full record whose linearization is70h-7Fh, call Get Sensor Reading Factors with the raw reading obtained from Get Sensor Reading, and convert with the returned factors. The thresholds of such a record would need the same treatment (one call per threshold byte) if they are to be reported.hasAnalogReading()then staysfalseonly for the data format11b.Evidence: by the specification. Neither test BMC (GIGABYTE, Lenovo IMM) has a non-linear sensor, so the implementation can only be verified with unit tests on crafted bytes; say so in the PR.
Related: #83 (closed by #143 with the fail-safe behaviour), #111 and #101 (other client API enhancements). Low priority: non-linear sensors are rare in practice.