SNAS264D April 2006 – February 2024 LM94
PRODUCTION DATA
PROCHOT monitoring applies to both the P1_PROCHOT and P2_PROCHOT inputs. Both inputs are monitored in the same fashion, but the following description discusses a single monitor. ( Px_PROCHOT represents both P1_PROCHOT and P2_PROCHOT).
PROCHOT monitoring is meant to achieve two goals. One goal is to measure the percentage of time that PROCHOT is asserted over a programmable time period. The result of this measurement can be read from an 8-bit register where one LSB equals 1/256th of the PROCHOT Time Interval (0.39%). The second goal is to have a status register that indicates, as a coarse percentage, the amount of time a processor has been throttled. This second goal is required in order to communicate information over the NIC using ASF, i.e. status can be sent, not values.
To achieve the first goal, the PROCHOT input is monitored over a period of time as defined by the PROCHOT Time Interval Register. At the end of each time period, the 8-bit measurement is transferred to the Current Px_PROCHOT register. Also at the end of each measurement period, the Current Px_PROCHOT register value is moved to the Average Px_PROCHOT register by adding the new value to the old value and dividing the result by 2. Note that the value that is averaged into the Average Px_PROCHOT register is not the new measurement but rather the previous measurement. If the SMBus writes to the Current P1_PROCHOT (or Current P2_PROCHOT) register, the capture cycle restarts for both monitoring channels ( P1_PROCHOT and P2_PROCHOT). Also note, that a strict average of two 8-bit values may result in Average Px_PROCHOT reflecting a value that is one LSB lower than the Current Px_PROCHOT in steady state.
It should be noted that the 8-bit result has a positive bias of one half of an LSB. This is necessary because a value of 00h represents that Px_PROCHOT was not asserted at all during the sampling window. Any amount of throttling results in a reading of 01h.
The following table demonstrates the mapping for the 8-bit result:
| 8–Bit Result | Percentage Thottled |
|---|---|
| 0 | Exactly 0% |
| 1 | Between 0% and 0.39% |
| 2 | Between 0.39% and 0.78% |
| - | - |
| n | Between (n-1)/256 and n/256 |
| - | - |
| 253 | Between 98.4% and 98.8% |
| 254 | Between 98.8% and 99.2% |
| 255 | Greater than 99.2% |
To achieve the second goal, the LM94 has several comparators that compare the measured percentage reading against several fixed and 1 variable value. The variable value is user programmable.
The result of these comparisons generates several error status bits described in the following table:
| Status Description | Comparison Formula |
|---|---|
| 100% Throttle | PROCHOT was never de-asserted during monitoring interval. |
| Greater than or equal to 75% and less than 100% | 193 ≤ measured value and not 100% |
| Greater than or equal to 50% and less than 75% | 129 ≤ measured value < 193 |
| Greater than or equal to 25% and less than 50% | 65 ≤ measured value < 129 |
| Greater than or equal to 12.5% and less than 25% | 33 ≤ measured value < 65 |
| Greater than 0% and less than 12.5% | 0 < measured value < 33 |
| Greater than 0% | 0 < measured value |
| Greater than user limit | user limit < measured value |
These status bits are reflected in the PROCHOT Error Status Registers. Each of the P1_PROCHOT and P2_PROCHOT inputs is monitored independently, and each has its own set of status registers.
In S3 and S4/5 sleep states, the PROCHOT Monitoring function does not run. VRDx_Hot is disabled from activating PROCHOT pins in S3 and S4/5. The Current Px_PROCHOT registers are reset to 00h and the Average Px_PROCHOT registers hold their current state. Once the sleep state changes back to S0, the monitoring function is restarted. After the first PROCHOT measurement has been made, the measurement is written directly into the Current and Average Px_PROCHOT registers without performing any averaging. Averaging returns to normal on the second measurement.