A smart power supply method and device that combines dynamic calibration and fault tracing functions

By configuring the PMU in the baseboard management controller (BMC), power operation data of the power rail is collected and stored, fault tracing results are generated, and power supply parameters are dynamically calibrated. This solves the problem of difficulty in tracing back the power supply data before the fault after a memory error event, and achieves fault tracing and calibration with high accuracy and safety.

CN122816432BActive Publication Date: 2026-11-10KINGTIGER TESTING TECH (SZ) LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611322920.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-10
Estimated Expiration
2046-08-28

AI Technical Summary

Technical Problem

After a memory error occurs, existing technologies struggle to preserve and trace back transient power supply data prior to the fault, making fault tracing and dynamic calibration difficult. In particular, transient power supply changes are not easily preserved in high-load operating scenarios.

Method used

By configuring a power management unit (PMU) in the baseboard management controller (BMC), power operation data of the power rail is collected and stored. After a memory error occurs, the data within the target time window is frozen. Fault tracing results are generated by combining preset mapping relationships and abnormal timing, and power supply parameters are dynamically calibrated.

Benefits of technology

It enables accurate identification of the correlation between power supply anomalies after memory error events, improving the accuracy of fault tracing, and ensures the safety and stability of dynamic calibration through hierarchical step calibration and protection mechanisms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122816432B_ABST
    Figure CN122816432B_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent power supply method and apparatus that combines dynamic calibration and fault tracing functions, relating to the field of intelligent power supply technology. The Baseboard Management Controller (BMC) receives error event information, including memory location information, sent by the BIOS or firmware based on memory errors. Based on the error event information, it determines the target power management unit, target power rail, and target time window, and sends a freeze query request to the target power management unit via I2C, I3C, or SPI. The target power management unit freezes the cached data corresponding to the target time window or copies the cached data to a protected area and returns the corresponding power operation data. The BMC generates fault tracing results based on anomalies in input voltage, output voltage, output current, and output power consumption and their timing relationships, and sends a dynamic calibration command when memory errors are related to power supply anomalies. This invention can preserve transient power supply data before a fault, achieving fault tracing and safe dynamic calibration of the target power supply domain.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent power supply technology, and in particular to an intelligent power supply method and device that combines dynamic calibration and fault tracing functions. Background Technology

[0002] With the widespread application of computing systems such as servers, high-performance computing devices, and artificial intelligence training equipment, memory modules require continuous and stable power supply support under high-load operating scenarios. During memory operation, abnormal situations such as correctable errors, uncorrectable errors, or intermittent errors may occur. These abnormalities may be influenced by a combination of factors, including the memory device itself, its contact status, the software operating environment, the processor's memory controller status, and the power supply status. In actual operation and maintenance, when computing devices experience blue screens, crashes, or memory error events, a comprehensive judgment is usually required, combining system logs, error codes, hardware status records, and manual inspection results.

[0003] In existing computing devices, the operating system can generate dump files for software analysis after an anomaly, and the baseboard management controller can monitor and record platform operating parameters such as voltage, temperature, fan speed, and power supply status. These methods can provide some basis for troubleshooting. However, in scenarios where memory errors are related to transient power supply changes, changes in power rail voltage, current, or power consumption in the very short time before the fault occur are not easily preserved completely, making it difficult for maintenance personnel to accurately determine whether the memory error is related to the power supply anomaly. Especially in scenarios with rapidly changing loads such as memory stress testing, big data processing, and artificial intelligence training, relying solely on low-frequency inspection data or error codes after the fault makes it difficult to reconstruct the transient power supply operating state before the fault. Therefore, it is necessary to provide an intelligent power supply method and device that can trace back power supply operating data before the fault after a memory error event occurs, and perform fault tracing and dynamic calibration based on this data. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent power supply method and device that combines dynamic calibration and fault tracing functions to solve the problems of difficulty in preserving and tracing back transient power supply data before the fault after a memory error event occurs, and difficulty in performing fault tracing and safety dynamic calibration based on the power supply data.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent power supply method that combines dynamic calibration and fault tracing functions, applied to a baseboard management controller (BMC) of a computing device. The computing device includes a processor, memory, and a power system. The power system includes multiple power management units (PMUs) for providing power to the memory. Each PMU is configured to collect and store power operation data for a corresponding power rail according to a time identifier. The method comprises:

[0008] Receive error event information sent by the interrupt handler of the BIOS or firmware based on a memory error detected by the processor, the error event information including memory location information;

[0009] Based on the error event information and the preset mapping relationship, the target time, target time window, target PMU, and target power rail are determined. The target time window includes a preset time period before the target time, and the preset mapping relationship is used to characterize the correspondence between memory location, PMU, and power rail.

[0010] A freeze query request is sent to the target PMU via the I2C, I3C, or SPI management bus. The freeze query request is used to cause the target PMU to freeze the cached data corresponding to the target time window or copy the cached data to the protected area, and return the power operation data of the target power rail within the target time window.

[0011] Receive the power operation data returned by the target PMU;

[0012] Based on the power supply operation data and the occurrence sequence of each anomaly in the power supply operation data, a fault tracing result is generated to characterize the correlation between the memory error and the power supply anomaly.

[0013] If the fault tracing results indicate that the memory error is related to a power supply anomaly, a dynamic calibration instruction is generated based on the fault tracing results, and the dynamic calibration instruction is sent to the target PMU so that the target PMU adjusts the power supply parameters of the target power rail.

[0014] As a preferred embodiment of the method described in this invention, the power supply operation data includes the input voltage of the target PMU, and at least one of the output voltage, output current, output power consumption, temperature, and status flags of the target power supply rail.

[0015] In a preferred embodiment of the method described in this invention, each of the PMUs is configured to collect and store power operation data for the corresponding power rail according to a time identifier, including:

[0016] The collected power supply operation data is written into a circular buffer.

[0017] When the circular buffer is full, the earliest power operation data written in the circular buffer is overwritten with the newly acquired power operation data.

[0018] In a preferred embodiment of the method described in this invention, receiving error event information sent by the interrupt handler of the BIOS or firmware based on a memory error detected by the processor includes:

[0019] The processor detects the memory error based on ECC verification and triggers a hardware interrupt.

[0020] In response to the hardware interrupt, the interrupt handler of the BIOS or firmware reads the processor machine check register and / or memory controller error register to obtain at least one of the following: error type, error occurrence time, memory address, memory channel identifier, and memory module identifier, and sends error event information including the memory location information to the BMC.

[0021] The preset mapping relationship includes the correspondence between processor identifier, memory controller identifier, memory address range, memory channel identifier, memory module identifier, PMU identifier, and power rail identifier; the target PMU is the PMU that corresponds to the memory location information among the multiple PMUs included in the power system.

[0022] In a preferred embodiment of the method described in this invention, the step of generating a fault tracing result characterizing the correlation between the memory error and the power supply abnormality based on the power supply operation data and the occurrence sequence of each abnormality in the power supply operation data includes:

[0023] If the input voltage of multiple PMUs powered by the same upstream power supply drops simultaneously within the first time range, and the output voltage of the corresponding multiple power supply rails subsequently drops, the fault source type is determined to be an abnormal upstream power supply input.

[0024] If the input voltage of the target PMU is within a preset input voltage range, the output voltage of the target power supply rail is abnormal, and no significant increase in output current or output power consumption was detected before the output voltage abnormality, the fault source type is determined to be a power management unit malfunction.

[0025] If the output current or output power consumption of the target power rail increases before the output voltage decreases, the fault source type is determined to be an abnormal memory load change.

[0026] If the output current of the power rail corresponding to a single memory module increases abnormally and the error events are concentrated in the single memory module, the fault source type is determined to be a memory module malfunction.

[0027] If no abnormal input voltage, abnormal output voltage, sudden output current, or sudden output power consumption is detected within the target time window, the fault source type is determined to be a non-power supply related anomaly.

[0028] As a preferred embodiment of the method described in this invention, the step of generating dynamic calibration instructions based on the fault tracing results includes:

[0029] The target power supply rail to be adjusted and the adjustment amount of the target power supply rail are determined based on the fault tracing results.

[0030] Within a preset safety range, a dynamic calibration command is generated, including the target power supply rail, the adjustment amount, and the adjustment direction.

[0031] As a preferred embodiment of the method described in this invention, the generation of dynamic calibration instructions includes:

[0032] Calculate the deviation between the abnormal voltage value of the target power supply rail and the preset normal voltage range;

[0033] If the deviation value is greater than the first deviation threshold, a coarse adjustment command including a first preset adjustment step is generated;

[0034] If the deviation value is greater than the second deviation threshold and less than or equal to the first deviation threshold, a fine-tuning instruction including a second preset adjustment step is generated.

[0035] If the deviation value is less than or equal to the second deviation threshold, a fine-tuning command or a stop-adjustment command is generated;

[0036] Wherein, the first preset adjustment step is greater than the second preset adjustment step.

[0037] As a preferred embodiment of the method described in this invention, after sending the dynamic calibration command to the target PMU, the method further includes:

[0038] After a preset stabilization time, the adjusted output voltage is obtained;

[0039] If the adjusted output voltage is within the preset normal voltage range, continue monitoring a preset number of sampling points and then end the calibration.

[0040] If the adjusted output voltage is not within the preset normal voltage range, the deviation value is recalculated and calibration continues.

[0041] As a preferred embodiment of the method described in this invention, the method further includes:

[0042] If the number of cycles in a single calibration process reaches a preset number and the output voltage is still not within the preset normal voltage range, the adjustment is stopped and a power hardware abnormality alarm is generated.

[0043] If two consecutive adjustments are made in opposite directions, the adjustment will be stopped and a power supply oscillation alarm will be generated.

[0044] If the adjusted power supply operating data triggers a preset protection condition, or if an error event information corresponding to the target power supply rail is still received after adjustment, a rollback command is sent to the target PMU to restore the target PMU to the power supply parameters before adjustment.

[0045] Secondly, the present invention provides an intelligent power supply device that combines dynamic calibration and fault tracing functions, applied to a baseboard management controller (BMC) of a computing device. The computing device includes a processor, memory, and a power system. The power system includes multiple power management units (PMUs) for providing power to the memory. The intelligent power supply device comprises:

[0046] The event receiving unit is configured to receive error event information, including memory location information, sent by the interrupt handler of the BIOS or firmware.

[0047] The target determination unit is configured to determine the target time, target time window, target PMU, and target power supply rail based on the error event information and a preset mapping relationship.

[0048] The freeze query unit is configured to send a freeze query request to the target PMU via the I2C, I3C or SPI management bus, and receive the power operation data of the target power rail within the target time window returned by the target PMU.

[0049] The fault tracing unit is configured to generate fault tracing results based on the power supply operation data and the occurrence sequence of each abnormality in the power supply operation data;

[0050] The dynamic calibration unit is configured to generate a dynamic calibration instruction based on the fault tracing result when the fault tracing result indicates that the memory error is related to the power supply abnormality, and send the dynamic calibration instruction to the target PMU;

[0051] The monitoring and protection unit is configured to monitor the adjusted power supply operation data and subsequent error event information, and to stop the adjustment, generate alarms and / or send rollback commands when preset protection conditions are triggered.

[0052] The beneficial effects of this invention are as follows: the BMC can directly locate the target PMU and the target power rail based on the memory location information reported once, and preserve the transient power supply data within the target time window before the fault by freezing the query request; based on the anomalies and sequence of input voltage, output voltage, output current and output power consumption, the accuracy of fault source judgment can be improved; through graded step calibration, cycle number limitation, oscillation detection and backoff mechanism, the safety and stability of dynamic calibration can be improved. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic diagram of the intelligent power system architecture and management bus connection that combines dynamic calibration and fault tracing functions, provided in an embodiment of the present invention.

[0055] Figure 2 A schematic diagram of the intelligent power supply method with dynamic calibration and fault tracing functions provided in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram illustrating the hardware interrupt reporting, cache freezing, target time window query, and power rail mapping relationship provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0058] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0059] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0060] Reference Figures 1-3This embodiment of the invention provides an intelligent power supply method that combines dynamic calibration and fault tracing functions. The method is applied to the BMC (Body Control Center) of a computing device, which includes a processor, memory, and a power system. The power system includes one or more PMUs (Power Management Units), each PMU including an MCU (Microcontroller Unit), a DC-DC power conversion unit, a high-precision sampling and interface control unit (HPSIC), a storage unit, and a communication interface unit. Figure 1 As shown, the BMC can communicate with the CPU via IPMI or IPMB, and the BMC can communicate with each PMU via one or more management buses such as I2C, I3C, and SPI. The HPSIC is used to collect the voltage, current, power consumption, and status flags of the power supply rails, and can work with the MCU to perform data buffering, freezing, and return.

[0061] In one specific implementation, one PMU corresponds to one memory power domain, or one PMU corresponds to one memory channel, one memory module, or a group of memory modules. The power rails corresponding to the memory include at least one of VDD, VDDQ, and VPP. VDD is used to provide power to the memory core circuitry, VDDQ is used to provide power to the memory input / output circuitry, and VPP is used to provide power to the memory word line driver-related circuitry. In practical applications, the names and number of power rails can be adjusted according to the memory type and platform power supply architecture.

[0062] Specifically, the intelligent power supply method that combines dynamic calibration and fault tracing includes the following steps.

[0063] Step S1: The power system collects the power operation data of the corresponding power rail in the memory in real time, and stores the power operation data in association with the corresponding time stamp.

[0064] During the computing device initialization phase, the BMC sends time synchronization information to each PMU. Each PMU records its local counter value when it receives the time synchronization information. The BMC establishes an offset relationship between the BMC time and the local counters of each PMU. When storing power operation data, the PMU records its local counter value. After determining the target time window, the BMC converts the target time window into the local counter range of the target PMU based on the offset relationship. In an implementation without time synchronization, the BMC determines the corresponding buffered data range based on the error event reception time, the preset event reporting delay, and the PMU sampling period.

[0065] After the computing device powers on or the BMC initializes, the BMC sends a sampling configuration to the power system. The sampling configuration includes the sampling object, sampling period, cache depth, and data type. The sampling object is one or more memory power rails, such as VDD, VDDQ, and VPP. The sampling period is a pre-set period, preferably shorter than the BMC's regular inspection period. For example, the BMC's regular inspection period can be 1 to 10 seconds, while the power system's local sampling period can be 10 microseconds to 10 milliseconds. The specific sampling period can be determined based on the sampling chip's performance, communication bandwidth, and cache capacity.

[0066] Power supply operating data includes the PMU input voltage, and at least one of the following: output voltage, output current, and output power consumption of the power supply rail. To further improve fault tracing capabilities, power supply operating data may also include at least one of the following: input current, temperature, target voltage, duty cycle, overvoltage indicator, undervoltage indicator, overcurrent indicator, overtemperature indicator, and power supply good status indicator.

[0067] After acquiring power operation data, the power system associates and stores the data with corresponding timestamps. These timestamps can be PMU local counter values, BMC synchronization timestamps, sampling sequence numbers, or system time converted from local counters. To reduce storage costs, the power system preferably uses a circular cache to store power operation data. The circular cache duration is longer than the maximum preset delay between the occurrence of a memory error and the target PMU receiving a freeze query request. Specifically, the power system sequentially writes the acquired power operation data into the circular cache; when the circular cache is full, the oldest written power operation data is overwritten with newly acquired power operation data, ensuring that the circular cache always retains power operation data from the most recent period.

[0068] For example, a circular buffer is configured to store the most recent 60 seconds of power supply operation data. When the sampling time exceeds 60 seconds, data collected in the 61st second overwrites data collected in the 1st second, data collected in the 62nd second overwrites data collected in the 2nd second, and so on. Each data record in the circular buffer may include the following fields: PMU identifier, power rail identifier, sampling time identifier, output voltage, output current, output power consumption, and status flag.

[0069] Step S2: The BMC receives error event information generated by the processor based on memory errors.

[0070] During the operation of the computing device, the processor checks memory read and write data using ECC (Electronic Code Correction). When a memory error is detected, the processor triggers a hardware interrupt, which can be an SMI (Signal Mitigation Error), NMI (Non-Signal Mitigation Error), or other platform-defined interrupts triggered by an MCA (Mean Change Detection) exception. Error event information may include correctable error event information CE-error and / or uncorrectable error event information UCE-error.

[0071] In response to the hardware interrupt, the interrupt handler in the BIOS or firmware reads the processor machine check register, memory controller error register, or other platform error register to obtain at least one of the following: error type, error occurrence time, memory address, memory channel identifier, memory module identifier, rank information, and bank information. The error event information is then sent to the BMC via IPMB, I2C, SMBus, KCS, or a platform-defined management channel.

[0072] In this embodiment, the error event information already includes the address, channel, or memory module information for locating the faulty memory in a single report. Therefore, after receiving the error event information, the BMC can determine the target PMU and the target power rail without waiting for subsequent supplementary memory location information.

[0073] Step S3: The BMC determines the target power management unit and the target power rail based on the error event information.

[0074] The BMC pre-maintains the mapping relationship between memory locations and PMUs and power rails. This mapping relationship can be provided by the BIOS, platform configuration table, FRU information, or preset configuration file during system initialization, and includes the correspondence between processor identifiers, memory controller identifiers, memory address ranges, memory channel identifiers, memory module identifiers, PMU identifiers, and power rail identifiers.

[0075] For example, CPU0's first memory channel corresponds to PMU1, and CPU0's second memory channel corresponds to PMU2; each PMU outputs VDD, VDDQ, and VPP to its corresponding memory channel. When the BMC receives an error event information, it uses the memory address, memory channel identifier, or memory module identifier to find the target PMU and target power rail corresponding to the memory error in the mapping relationship.

[0076] Since the error event information already includes memory location information, the BMC can send a freeze query request only to the target PMU, thereby avoiding the data processing and communication overhead caused by freezing all PMUs.

[0077] Step S4: The BMC sends a freeze query request to the target PMU and obtains the power operation data within the target time window.

[0078] The BMC determines the target time and target time window based on the error event information. The target time is preferably the time when the memory error occurred or the time when the processor detected it, carried in the error event information. When the error event information does not carry a corresponding time, the target time can also be the time when the BMC receives the error event information, or a time determined based on the receiving time and a preset transmission delay compensation amount. The target time window includes a preset time period before the target time, such as 100 milliseconds, 1 second, 10 seconds, or 60 seconds before the target time.

[0079] The BMC sends a freeze query request to the target PMU via the I2C, I3C, or SPI management bus. The freeze query request may include an error event identifier, a target power rail identifier, the start and end times of the target time window, and the data type requested.

[0080] Upon receiving a freeze query request, the target PMU stops overwriting the circular buffer segment corresponding to the target time window, or copies the data in the buffer segment to an independent protected area. Subsequently, it determines the power operation data of the target power rail within the target time window based on the time identifier and returns this power operation data to the BMC. While freezing the corresponding buffer segment, the target PMU can continue to write newly acquired data to other buffer segments to balance pre-fault data preservation and post-fault status monitoring.

[0081] In one implementation, the BMC first requests summary data within the target time window, which includes the minimum input voltage, the minimum and maximum output voltage, the maximum output current, the maximum output power consumption, and an abnormal status flag. If the summary data indicates an abnormality, the BMC then requests high-frequency raw sampling data before and after the time the abnormality occurred.

[0082] Step S5: The BMC generates fault tracing results based on the power supply operation data within the target time window.

[0083] The BMC detects whether the input voltage, output voltage, output current, and output power consumption of the target PMU are abnormal, and determines the timing and sequence of each abnormality. The preset voltage threshold can be determined based on the memory device specifications, platform power supply specifications, or factory calibration results; the preset current change threshold and preset power consumption change threshold can be determined based on the results of normal load switching tests with a safety margin.

[0084] When at least one anomaly occurs within a preset relevant time range prior to the target time, the BMC determines that the memory error is related to a power supply anomaly. The preset relevant time range can be set based on the PMU sampling period, processor error detection latency, and the transmission latency of the BIOS or firmware reporting error event information to the BMC; for example, it can be set to 10 microseconds to 1 second prior to the target time.

[0085] When the input voltage of multiple PMUs powered by the same upstream power supply drops simultaneously within a first time range, and the output voltage of the corresponding multiple power supply rails subsequently drops, the BMC determines that the fault source type is an abnormal upstream power supply input.

[0086] When the input voltage of the target PMU is within the preset input voltage range, the output voltage of the target power supply rail is abnormal, and no significant increase in output current or output power consumption was detected before the output voltage abnormality, the BMC determines the fault source type as a power management unit malfunction.

[0087] When the output current or output power consumption of the target power rail rises rapidly and then the output voltage drops, the BMC determines that the fault source is an abnormal memory load change.

[0088] When the output current of the power rail corresponding to a single memory module increases abnormally and the error events are concentrated in that memory module, the BMC determines that the fault source type is a memory module malfunction.

[0089] If no abnormal input voltage, output voltage, output current, or output power consumption is detected within the target time window, the BMC determines the fault source type as a non-power supply related anomaly. By combining the anomaly magnitude, duration, and sequence of occurrence, false positives caused by relying solely on a single threshold for fault diagnosis can be reduced.

[0090] It should be noted that when the same error event simultaneously meets the judgment conditions of multiple fault source types, BMC will write the multiple fault source types that meet the judgment conditions as candidate fault source types into the fault tracing results, and record the abnormal data and the sequence of occurrence of each candidate fault source type.

[0091] Step S6: The BMC generates a dynamic calibration command based on the fault tracing results and sends it to the power system.

[0092] When the fault tracing results indicate that memory errors are related to power supply anomalies, the BMC determines the target power rail to be adjusted, the adjustment direction, and the adjustment amount based on the fault source type, and generates dynamic calibration instructions within the safe range specified by the memory device specifications and platform power supply specifications. Power supply parameters may include at least one of the following: output voltage, load line parameters, transient response compensation parameters, switching frequency, power supply phase state, and current limiting threshold.

[0093] In one implementation, the BMC calculates the deviation between the abnormal voltage value of the target power supply rail and a preset normal voltage range. When the deviation value is greater than a first deviation threshold, a coarse adjustment command including a first preset adjustment step is generated; when the deviation value is greater than a second deviation threshold but less than or equal to the first deviation threshold, a fine adjustment command including a second preset adjustment step is generated; when the deviation value is less than or equal to the second deviation threshold, a fine adjustment command or a stop adjustment command is generated. The first preset adjustment step is greater than the second preset adjustment step.

[0094] Taking a VDDQ power supply rail with a nominal voltage of 1.10V as an example, the preset normal voltage range can be 1.08V to 1.12V; the coarse adjustment step, fine adjustment step, and micro adjustment step can be set to 20mV, 10mV, and 5mV, respectively. The above values ​​are for illustrative purposes only, and the actual values ​​can be determined based on the memory device specifications, PMU adjustment resolution, and platform test results.

[0095] The BMC sends a dynamic calibration command to the target PMU. The target PMU adjusts the target power rail step by step according to the adjustment steps and direction in the dynamic calibration command, and waits for a preset stabilization time after adjustment. The preset stabilization time is, for example, 20ms, but can also be set according to the PMU response time and the power rail type.

[0096] After a preset stabilization time, the BMC reads the adjusted output voltage. If the adjusted output voltage is within the preset normal voltage range, the calibration ends, and monitoring continues for 3 to 5 sampling points; if it is still not within the preset normal voltage range, the deviation value is recalculated and adjustment continues.

[0097] Step S7: BMC verifies the dynamic calibration results and rolls back in case of anomalies.

[0098] After the target PMU performs dynamic calibration, it returns the adjusted power operation data and adjustment status to the BMC. The BMC continues to monitor the power operation data and subsequent error event information corresponding to the target memory location or target power rail.

[0099] If the adjusted output voltage is within the preset normal voltage range, and the number of similar error events decreases during the observation period, the BMC determines that the dynamic calibration is effective and retains the adjusted power supply parameters.

[0100] When the number of cycles in a single calibration process reaches the preset number, such as 3 times, and the output voltage is still not stable, the BMC stops adjusting and generates a power supply hardware abnormality alarm; when two consecutive adjustments are in opposite directions, the BMC determines that there may be power supply oscillation or abnormal control direction, stops adjusting and generates a power supply oscillation alarm.

[0101] When the adjusted power supply operating data triggers a preset protection condition, or when an error event corresponding to the target power supply rail is still received after adjustment, the BMC sends a rollback command to the target PMU to restore the target PMU to the power supply parameters before adjustment. The preset protection condition includes at least one of the following: output voltage exceeding the preset voltage limit, output current exceeding the preset current limit, temperature exceeding the preset temperature threshold, overcurrent flag triggering, overtemperature flag triggering, or power supply good condition failure.

[0102] The aforementioned cycle count limit, adjustment direction reversal detection, and rollback mechanism can reduce the risks caused by over-adjustment, repeated oscillation, or continuous adjustment in the event of power supply hardware malfunction.

[0103] Step S8: BMC generates fault tracing records.

[0104] After completing fault tracing and dynamic calibration, BMC generates a fault tracing record. The fault tracing record includes at least one of the following: error event identifier, error type, memory address, memory channel identifier, memory module identifier, target PMU identifier, target power rail identifier, target time window, minimum voltage value, maximum current value, maximum power consumption value, time of anomaly occurrence, fault source type, dynamic calibration instruction, and calibration result.

[0105] The fault tracing record can be stored in the BMC log, system event log, or remote operation and maintenance platform for subsequent analysis and maintenance by operation and maintenance personnel.

[0106] This embodiment also provides an intelligent power supply device that combines dynamic calibration and fault tracing functions. The device is applied to the BMC of a computing device and includes an event receiving unit, a target determination unit, a freeze query unit, a fault tracing unit, a dynamic calibration unit, and a monitoring and protection unit.

[0107] The event receiving unit is configured to receive error event information, including memory location information, sent by the interrupt handler of the BIOS or firmware; the target determination unit is configured to determine the target time, target time window, target PMU, and target power rail based on the error event information and a preset mapping relationship.

[0108] The freeze query unit is configured to send a freeze query request to the target PMU via the I2C, I3C, or SPI management bus, and to receive power operation data within the target time window returned by the target PMU from the freeze buffer segment or protected area.

[0109] The fault tracing unit is configured to generate fault tracing results based on the anomalies of input voltage, output voltage, output current, and output power consumption and their order; the dynamic calibration unit is configured to generate dynamic calibration instructions and send them to the target PMU when memory errors are related to power supply anomalies.

[0110] The monitoring and protection unit is configured to monitor the adjusted power supply operation data and subsequent error event information, and to stop adjustment, generate alarms, and / or send rollback commands when protection conditions are triggered, the number of cycles reaches a limit, or a reversal of the adjustment direction is detected. The specific implementation of each unit in the device can be found in the above-described method embodiment.

[0111] Example 2

[0112] This embodiment provides a power data processing method for a target PMU, which is used in conjunction with the aforementioned BMC-side method.

[0113] The target PMU collects at least one of the following in real time: input voltage, output voltage, output current, output power consumption, and status flags of the corresponding power rail, and writes the power operation data into a circular buffer along with the corresponding time flag. When the circular buffer is full, the newly collected data overwrites the oldest written data.

[0114] The target PMU receives the freeze query request sent by the BMC via the I2C, I3C or SPI management bus, stops covering the corresponding buffer segment according to the target time window in the freeze query request, or copies the data of the corresponding buffer segment to the independent protection zone; then determines the target data according to the target power rail identifier and time identifier, and returns the target data to the BMC.

[0115] The target PMU receives the dynamic calibration command sent by the BMC, adjusts the power supply parameters of the target power supply rail according to the adjustment amount, adjustment direction and adjustment step in the dynamic calibration command, and returns the adjusted power operation data and adjustment status to the BMC; when a rollback command is received, the target power supply rail is restored to the power supply parameters before the dynamic calibration command was executed.

[0116] Example 3: To verify the beneficial effects of the present invention, a memory stress test task was run on a computing device in a specific application scenario. The PMU1 in the power supply system provides VDD, VDDQ, and VPP power to the first memory channel of CPU0. The PMU1 collects the input voltage, output voltage, output current, and output power consumption of the VDDQ power rail at a 1-millisecond sampling period, and writes the sampled data along with the PMU1's local time identifier into a circular buffer. The circular buffer stores data from the most recent 60 seconds.

[0117] At time T0, an uncorrectable error occurs in the first memory channel of CPU0. The processor detects this error via ECC check and triggers a hardware interrupt. The interrupt handler in the BIOS or firmware reads the relevant error register, generates error event information including the UCE-error type, the CPU0 first memory channel identifier, the corresponding memory module identifier, the memory address, and the time of the error occurrence, and sends the error event information to the BMC. Since the error event information already includes memory location information, the BMC can determine the target PMU without waiting for subsequent supplementary information.

[0118] The BMC uses the memory channel identifier and memory address to look up the preset mapping relationship, determines the target PMU as PMU1 and the target power rail as VDDQ, and uses 1 second before T0 as the target time window. The BMC sends a freeze query request to PMU1 via I2C, I3C, or SPI; PMU1 stops overwriting the ring buffer segment corresponding to the target time window or copies the data of that segment to the protected area, and then returns the input voltage, output voltage, output current, and output power consumption data of the VDDQ power rail within the target time window.

[0119] BMC analyzed the returned data and found that the PMU1 input voltage remained within the preset input voltage range. The VDDQ output current and output power consumption first rose rapidly, and then the VDDQ output voltage fell below the preset lower voltage limit within 20 milliseconds before T0. Based on this, BMC determined that the UCE-error event was related to an abnormal VDDQ power supply, and that the fault source was an abnormal sudden change in memory load.

[0120] Based on the fault tracing results, the BMC determines that the power supply rail to be adjusted is the VDDQ power supply rail of PMU1, and generates a graded step dynamic calibration command based on the deviation between the VDDQ output voltage and the preset normal voltage range. PMU1 gradually adjusts the VDDQ output voltage according to the dynamic calibration command, and waits for a preset stabilization time after each adjustment. If the adjusted output voltage is within the preset normal voltage range and is stable after continuous monitoring of 3 to 5 sampling points, the BMC records that the dynamic calibration is valid; if it is still not stable after 3 cycles of adjustment, the adjustment direction is opposite twice in a row, the output current triggers the protection condition, or the corresponding error event occurs again, the BMC stops the adjustment and generates an alarm, and sends a rollback command to PMU1 if necessary.

[0121] The above solution collects and caches power operation data of the memory power rail in real time at the PMU end. After the BMC receives an error event including memory location information, it directly locates the target PMU and freezes the data of the target time window before the fault, which can reduce the risk of critical transient data being overwritten. At the same time, through multi-parameter timing fault tracing, hierarchical step calibration, oscillation detection and backoff mechanism, the accuracy and safety of fault location and dynamic calibration are improved.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart power supply method with both dynamic calibration and fault tracing functions, applied to a baseboard management controller (BMC) of a computing device, the computing device including a processor, memory, and a power system, the power system including multiple power management units (PMUs) for providing power to the memory, each PMU being configured to collect and store power operation data of a corresponding power rail according to a time identifier, characterized in that, The method includes: Receive error event information sent by the interrupt handler of the BIOS or firmware based on a memory error detected by the processor, the error event information including memory location information; Based on the error event information and the preset mapping relationship, the target time, target time window, target PMU, and target power rail are determined. The target time window includes a preset time period before the target time, and the preset mapping relationship is used to characterize the correspondence between memory location, PMU, and power rail. A freeze query request is sent to the target PMU via the I2C, I3C, or SPI management bus. The freeze query request is used to cause the target PMU to freeze the cached data corresponding to the target time window or copy the cached data to the protected area, and return the power operation data of the target power rail within the target time window. Receive the power operation data returned by the target PMU; Based on the power supply operation data and the occurrence sequence of each anomaly in the power supply operation data, a fault tracing result is generated to characterize the correlation between the memory error and the power supply anomaly. If the fault tracing results indicate that the memory error is related to a power supply anomaly, a dynamic calibration instruction is generated based on the fault tracing results, and the dynamic calibration instruction is sent to the target PMU so that the target PMU adjusts the power supply parameters of the target power rail.

2. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 1, characterized in that: The power supply operation data includes the input voltage of the target PMU, and at least one of the output voltage, output current, output power consumption, temperature, and status flags of the target power supply rail.

3. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 1, characterized in that, Each of the PMUs is configured to collect and store power operation data for the corresponding power rail according to a time identifier, including: The collected power supply operation data is written into a circular buffer. When the circular buffer is full, the earliest power operation data written in the circular buffer is overwritten with the newly acquired power operation data.

4. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 1, characterized in that, The receipt of error event information sent by the BIOS or firmware interrupt handler based on a memory error detected by the processor includes: The processor detects the memory error based on ECC verification and triggers a hardware interrupt. In response to the hardware interrupt, the interrupt handler of the BIOS or firmware reads the processor machine check register and / or memory controller error register to obtain at least one of the following: error type, error occurrence time, memory address, memory channel identifier, and memory module identifier, and sends error event information including the memory location information to the BMC. The preset mapping relationship includes the correspondence between processor identifier, memory controller identifier, memory address range, memory channel identifier, memory module identifier, PMU identifier, and power rail identifier; the target PMU is the PMU that corresponds to the memory location information among the multiple PMUs included in the power system.

5. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 2, characterized in that, The step of generating fault tracing results to characterize the correlation between the memory error and the power supply abnormality based on the power supply operation data and the occurrence sequence of each abnormality in the power supply operation data includes: If the input voltage of multiple PMUs powered by the same upstream power supply drops simultaneously within the first time range, and the output voltage of the corresponding multiple power supply rails subsequently drops, the fault source type is determined to be an abnormal upstream power supply input. If the input voltage of the target PMU is within a preset input voltage range, the output voltage of the target power supply rail is abnormal, and no significant increase in output current or output power consumption was detected before the output voltage abnormality, the fault source type is determined to be a power management unit malfunction. If the output current or output power consumption of the target power rail increases before the output voltage decreases, the fault source type is determined to be an abnormal memory load change. If the output current of the power rail corresponding to a single memory module increases abnormally and the error events are concentrated in the single memory module, the fault source type is determined to be a memory module malfunction. If no abnormal input voltage, abnormal output voltage, sudden output current, or sudden output power consumption is detected within the target time window, the fault source type is determined to be a non-power supply related anomaly.

6. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 1, characterized in that, The step of generating dynamic calibration instructions based on the fault tracing results includes: The target power supply rail to be adjusted and the adjustment amount of the target power supply rail are determined based on the fault tracing results. Within a preset safety range, a dynamic calibration command is generated, including the target power supply rail, the adjustment amount, and the adjustment direction.

7. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 6, characterized in that, The step of generating dynamic calibration instructions based on the fault tracing results includes: Calculate the deviation between the abnormal voltage value of the target power supply rail and the preset normal voltage range; If the deviation value is greater than the first deviation threshold, a coarse adjustment command including a first preset adjustment step is generated; If the deviation value is greater than the second deviation threshold and less than or equal to the first deviation threshold, a fine-tuning instruction including a second preset adjustment step is generated. If the deviation value is less than or equal to the second deviation threshold, a fine-tuning command or a stop-adjustment command is generated; Wherein, the first preset adjustment step is greater than the second preset adjustment step.

8. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 7, characterized in that, After sending the dynamic calibration command to the target PMU, the method further includes: After a preset stabilization time, the adjusted output voltage is obtained; If the adjusted output voltage is within the preset normal voltage range, continue monitoring a preset number of sampling points and then end the calibration. If the adjusted output voltage is not within the preset normal voltage range, the deviation value is recalculated and calibration continues.

9. The intelligent power supply method with both dynamic calibration and fault tracing functions as described in claim 8, characterized in that, The method further includes: If the number of cycles in a single calibration process reaches a preset number and the output voltage is still not within the preset normal voltage range, the adjustment is stopped and a power hardware abnormality alarm is generated. If two consecutive adjustments are made in opposite directions, the adjustment will be stopped and a power supply oscillation alarm will be generated. If the adjusted power supply operating data triggers a preset protection condition, or if an error event information corresponding to the target power supply rail is still received after adjustment, a rollback command is sent to the target PMU to restore the target PMU to the power supply parameters before adjustment.

10. An intelligent power supply device with both dynamic calibration and fault tracing functions, applied to the baseboard management controller (BMC) of a computing device, the computing device including a processor, memory, and a power system, the power system including multiple power management units (PMUs) for providing power to the memory, characterized in that, The intelligent power supply device includes: The event receiving unit is configured to receive error event information, including memory location information, sent by the interrupt handler of the BIOS or firmware. The target determination unit is configured to determine the target time, target time window, target PMU, and target power supply rail based on the error event information and a preset mapping relationship. The freeze query unit is configured to send a freeze query request to the target PMU via the I2C, I3C or SPI management bus, and receive the power operation data of the target power rail within the target time window returned by the target PMU. The fault tracing unit is configured to generate fault tracing results based on the power supply operation data and the occurrence sequence of each abnormality in the power supply operation data; The dynamic calibration unit is configured to generate a dynamic calibration instruction based on the fault tracing result when the fault tracing result indicates that the memory error is related to the power supply abnormality, and send the dynamic calibration instruction to the target PMU; The monitoring and protection unit is configured to monitor the adjusted power supply operation data and subsequent error event information, and to stop the adjustment, generate alarms and / or send rollback commands when preset protection conditions are triggered.

Citation Information

Patent Citations

  • Data access method and device in response to power event

    CN113535459A

  • Memory power supply detection method and device, electronic equipment and storage medium

    CN116755538A