Memory training method and device, equipment, medium and program product

Abstract

The application provides a memory training method, device, equipment, medium and program product, which can be applied to the technical field of computers. The memory training method comprises the following steps: in response to receiving a training log representing a memory training failure transmitted by a basic input / output system, analyzing the training log, determining a target training stage and a training failure field in which the failure occurs, and the training failure field comprising a target memory configuration parameter and state information representing a parameter value setting failure of the target memory configuration parameter; taking the target training stage, the target memory configuration parameter and the state information as graph nodes respectively, and taking an association relationship between the plurality of graph nodes as an edge, to generate a target graph; training the target graph to obtain an adjustment strategy; and updating a parameter value of the memory configuration parameter stored in a shared memory according to the adjustment strategy, so that the basic input / output system reads from the shared memory and re-executes the memory training based on the updated parameter value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a memory training method, device, equipment, medium and program product. BACKGROUND

[0002] Memory training is a key process in the start-up process of a computer system. In the related art, when memory training fails, a lookup table is used to adjust memory configuration parameters according to a preset static parameter table.

[0003] In the process of implementing the present application, it is found that the static parameter table relied on by the lookup table is fixed and cannot adapt to different hardware platforms and dynamically changing operating environments, resulting in the need for multiple attempts and system restarts when training fails, and low memory training efficiency. SUMMARY

[0004] In view of the above problems, the present application provides a memory training method, device, equipment, medium and program product.

[0005] According to a first aspect of the present application, a memory training method is provided, applied to a management controller, comprising: in response to receiving a training log representing memory training failure transmitted by a basic input / output system, parsing the training log to determine a target training phase where the failure occurs and a training failure field describing the training failure state, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing failure of setting a parameter value of the target memory configuration parameter; taking the target training phase, the target memory configuration parameter and the state information as graph nodes respectively, and taking the association relationship between the plurality of graph nodes as edges, to generate a target graph; training the target graph to obtain an adjustment strategy for the parameter value of the target memory configuration parameter; and updating the parameter value of the memory configuration parameter stored in a shared memory according to the adjustment strategy, so that the basic input / output system reads from the shared memory and re-executes memory training based on the updated parameter value.

[0006] The second aspect of the present application provides a memory training method applied to a basic input / output system, comprising: reading parameter values of memory configuration parameters from a shared memory, wherein the shared memory is accessed by the basic input / output system and a management controller; performing memory training based on the parameter values; in the case of failure of memory training, transmitting training logs representing the failure of memory training to the management controller to trigger the management controller to perform the following operations: in response to receiving the training logs representing the failure of memory training transmitted by the basic input / output system, analyzing the training logs to determine a target training phase in which the failure occurs and a training failure field describing the state of the training failure, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing failure of setting the parameter value of the target memory configuration parameter; generating a target graph by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking the association relationship between the plurality of graph nodes as edges; training the target graph to obtain an adjustment strategy for the parameter value of the target memory configuration parameter; updating the parameter values of the memory configuration parameters stored in the shared memory according to the adjustment strategy, so that the basic input / output system reads the updated parameter values from the shared memory and re-performs memory training based on the updated parameter values; in response to the management controller completing the update of the parameter values in the shared memory, re-reading the updated parameter values from the shared memory; and re-performing memory training based on the updated parameter values.

[0007] The third aspect of the present application provides a memory training device, comprising: a log analysis module configured to analyze training logs representing the failure of memory training transmitted by a basic input / output system in response to receiving the training logs, to determine a target training phase in which the failure occurs and a training failure field describing the state of the training failure, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing failure of setting the parameter value of the target memory configuration parameter; a graph construction module configured to generate a target graph by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking the association relationship between the plurality of graph nodes as edges; a strategy generation module configured to train the target graph to obtain an adjustment strategy for the parameter value of the target memory configuration parameter; and a parameter update module configured to update the parameter values of the memory configuration parameters stored in a shared memory according to the adjustment strategy, so that the basic input / output system reads the updated parameter values from the shared memory and re-performs memory training based on the updated parameter values.

[0008] The fourth aspect of the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method.

[0009] The fifth aspect of the present application also provides a computer-readable storage medium having stored thereon a computer program or instructions, which, when executed by a processor, implement the steps of the method.

[0010] The sixth aspect of the present application also provides a computer program product comprising a computer program or instructions, which, when executed by a processor, implement the steps of the method.

[0011] According to the embodiments of the present application, by parsing the training log to determine the target training stage, the training failure field and the corresponding target memory configuration parameter after receiving the training log representing the memory training failure transmitted by the basic input / output system, and constructing a target graph to intuitively present the fault correlation, the targeted adjustment strategy is trained based on the target graph, and then the parameter value in the shared memory is updated accordingly, which not only realizes the rapid positioning of the memory training failure, but also realizes the precise adjustment of the parameter value by optimizing the memory configuration parameter, significantly improves the success rate of the memory training and the overall reliability of the system, reduces the system downtime caused by the failure, and improves the operation and maintenance efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0012] The above and other objects, features and advantages of the present application will become more apparent from the following description when taken in conjunction with the accompanying drawings, in which:

[0013] Figure 1 An application scenario diagram of the memory training method, device, equipment, medium and program product according to the embodiments of the present application is shown.

[0014] Figure 2 A flowchart of the memory training method applied to the management controller according to the embodiments of the present application is shown.

[0015] Figure 3 A flowchart of the memory training method applied to the management controller according to another embodiment of the present application is shown.

[0016] Figure 4 A flowchart of the memory training method applied to the basic input / output system according to the embodiments of the present application is shown.

[0017] Figure 5 An interaction flowchart of the management controller and the basic input / output system according to the embodiments of the present application is shown.

[0018] Figure 6A structure block diagram of a memory training device applied to a management controller according to an embodiment of the present application is shown.

[0019] Figure 7 A structure block diagram of a memory training device applied to a basic input output system according to an embodiment of the present application is shown.

[0020] Figure 8 A block diagram of an electronic device adapted to implement a memory training method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0021] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. It is to be understood, however, that these descriptions are merely exemplary and are intended to provide a thorough and complete disclosure of the embodiments of the present application, as defined by the appended claims. In the following detailed description of the embodiments of the present application, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to one skilled in the art that the embodiments of the present application can be practiced without these specific details. In other instances, well-known structures and functions have not been described in detail in order to avoid obscuring the concepts of the present application.

[0022] The terms used herein are merely used to describe specific embodiments and are not intended to limit the present application. The terms "include" and "have" and the like used herein indicate the presence of the described features, steps, operations, and / or components but do not preclude the presence or addition of one or more other features, steps, operations, or components.

[0023] All terms used herein, including technical and scientific terms, have the same meanings as those generally understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings consistent with the context of the present specification, and should not be interpreted in an idealized or overly formal way.

[0024] In the case of using expressions similar to "at least one of A, B, and C, etc.", it is generally to be interpreted as including one or more of the items enumerated in the list (e.g., "a system having at least one of A, B, and C" should include, but not be limited to, a system having A alone, a system having B alone, a system having C alone, a system having both A and B, a system having both A and C, a system having both B and C, and / or a system having A, B, and C, etc.).

[0025] In the technical solutions of the present application, the data involved (including but not limited to data for analysis, stored data, displayed data, etc.) are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure, and application of relevant data comply with relevant laws, regulations, and standards, take necessary security measures, do not violate public order and good customs, and provide corresponding operation portals for user selection authorization or rejection.

[0026] Embodiments of the present application provide a memory training method, applied to a management controller, the method comprising: in response to receiving a training log representing a memory training failure transmitted by a basic input / output system, parsing the training log to determine a target training phase in which the failure occurs and a training failure field used to describe a training failure state, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing a parameter value setting failure of the target memory configuration parameter; taking the target training phase, the target memory configuration parameter, and the state information as graph nodes respectively, and taking an association relationship between the plurality of graph nodes as an edge, to generate a target graph; training the target graph to obtain an adjustment strategy for a parameter value of the target memory configuration parameter; and updating the parameter value of the memory configuration parameter stored in a shared memory according to the adjustment strategy, so that the basic input / output system reads from the shared memory and re-executes the memory training based on the updated parameter value.

[0027] Figure 1 An application scenario diagram of the memory training method, apparatus, device, medium, and program product according to embodiments of the present application is shown.

[0028] As shown in the figure, the application scenario 100 according to this embodiment can include a basic input / output system 101 and a management controller 102. The basic input / output system 101 and the management controller 102 can interact with each other, and the management controller 102 includes a shared memory 1021. A user can interact with the management controller 102 through the basic input / output system 101 to implement memory-related configuration, training, and other operations (only as an example).

[0029] The basic input / output system 101 can be a system with bottom-layer hardware control functions, and the management controller 102 can be a device for managing and controlling hardware resources such as memory, including but not limited to a dedicated management chip, an embedded controller, and the like.

[0030] The shared memory 1021 can be used to efficiently share data such as memory configuration parameters, training logs, and other information between the basic input / output system 101 and the management controller 102. The basic input / output system 101 can send memory training-related logs, requests, and other data to the management controller 102, the management controller 102 can quickly obtain and process these data through the shared memory 1021, and feed back the generated parameter adjustment strategy, updated parameter value, and other information to the basic input / output system 101, so as to realize effective management and optimization of the memory.

[0031] The following will be based on Figure 1 the described scenario, through Figure 2 and Figure 3The memory training method applied to the management controller in the embodiment of the present application is described in detail.

[0032] Figure 2 A flow chart of the memory training method applied to the management controller in the embodiment of the present application is shown.

[0033] As shown in Figure 2 , the embodiment includes operation S210 to operation S240.

[0034] In operation S210, in response to receiving a training log representing a memory training failure transmitted by a basic input output system, the training log is parsed to determine a target training stage where the failure occurs and a training failure field describing a training failure state, wherein the memory training goes through a plurality of training stages in sequence, the plurality of training stages respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing a parameter value setting failure of the target memory configuration parameter.

[0035] In operation S220, a target graph is generated by taking the target training stage, the target memory configuration parameter and the state information as graph nodes respectively and taking the association relationship between the plurality of graph nodes as edges.

[0036] In operation S230, the target graph is trained to obtain an adjustment strategy for the parameter value of the target memory configuration parameter.

[0037] In operation S240, according to the adjustment strategy, the parameter value of the memory configuration parameter stored in the shared memory is updated, so that the basic input output system reads from the shared memory and re-executes the memory training based on the updated parameter value.

[0038] According to the embodiment of the present application, in the computer system startup process, the memory training is a key link to ensure stable communication between the memory controller and the memory module. In this process, the configuration parameters used by the management controller for memory training are dynamically adjusted, thereby improving the success rate of memory initialization and system reliability. The management controller can be a baseboard management controller (BMC) or an integrated light-out 2 (ILO). In the specific implementation process, the management controller continuously listens to the transmission data from the basic input output system (BIOS), and when receiving a training log representing a memory training failure, the log parsing process is immediately performed.

[0039] In the process of analysis, key identification information is extracted from the training log to determine the target training phase corresponding to the training failure. Since memory training needs to go through multiple training phases in a predetermined order, and each training phase is responsible for adjusting a specific type of memory configuration parameter (such as timing parameter, voltage parameter, signal gain parameter, etc.), it is necessary to accurately locate the specific training phase where the fault occurs.

[0040] At the same time, the training failure field will also be selected from the training log. The training failure field not only includes the name and current parameter value of the target memory configuration parameter that caused the failure, but also covers state information representing the failure of the parameter value setting, such as parameter value exceeding the allowed range, parameter configuration incompatible with hardware, parameter taking effect timeout, and other specific fault descriptions.

[0041] After log analysis is complete, the management controller will build a target graph based on the analysis results. Specifically, the determined target training phase, target memory configuration parameter, and corresponding state information are used as three independent graph nodes, and each graph node will be attached with its own attribute information (such as the target training phase node including phase number, corresponding adjustment parameter type, the target memory configuration parameter node including parameter identification, default value range, and the state information node including fault code, fault occurrence time).

[0042] Subsequently, the edges of the graph are defined according to the logical association relationship between elements, such as establishing the association edge between the target training phase and the target memory configuration parameter, representing that the core task of the target training phase is to adjust the target memory configuration parameter, and establishing the association edge between the target memory configuration parameter and the state information, explicitly indicating that the configuration problem of the target memory configuration parameter directly leads to the corresponding failure state, thereby forming a target graph with complete structure and clear association.

[0043] The management controller will conduct targeted training on the generated target graph, and in the training process, it will call the built-in analysis model, combine the graph data accumulated from historical memory training failure cases, and mine the deep association rules between the graph nodes in the current target graph.

[0044] For example, the analysis of the same training phase, the same memory configuration parameter in different failure state parameter adjustment experience, learn the corresponding relationship between parameter value adjustment and fault repair, combined with the real-time state data of the current hardware environment (such as the current temperature, hardware load), optimize the model iteration, and finally output the precise adjustment strategy for the target memory configuration parameter. The adjustment strategy will clearly indicate the adjustment direction of the parameter value (such as up, down, keep fine tuning within the range), the specific adjustment amplitude, the verification standard after adjustment, and whether to adjust other related parameters. The adjustment strategy corresponding to different training phases and memory configuration parameters can be referred to Table 1, which shows the default values, training failure fields and optimization strategies of various memory configuration parameters in the timing training, signal optimization and voltage adaptation stages.

[0045] Table 1

[0046]

[0047] The management controller will access the shared memory with the basic input output system according to the generated adjustment strategy, find the storage address corresponding to the target memory configuration parameter, update the original parameter value stored in the storage address to the updated parameter value specified by the adjustment strategy, ensure the atomicity and consistency of the parameter update operation, and avoid data conflicts.

[0048] After the update is completed, the management controller will feed back the signal of parameter update completion to the basic input output system. After receiving the signal, the basic input output system will read the updated target memory configuration parameter value from the shared memory, and then re-execute the memory training, so as to improve the fault repair efficiency and the overall success rate of memory training.

[0049] By receiving the training log representing the memory training failure transmitted by the basic input output system, the log is analyzed to determine the failed target training phase, training failure field and corresponding target memory configuration parameter, and the target graph is constructed to intuitively present the fault association relationship. Based on the target graph, the targeted adjustment strategy is trained, and then the parameter value in the shared memory is updated, which not only realizes the rapid positioning of memory training failure, but also realizes the precise adjustment of parameter value by optimizing the memory configuration parameter, significantly improves the success rate of memory training and the overall reliability of the system, reduces the system downtime caused by failure, and improves the operation and maintenance efficiency.

[0050] According to the embodiment of the application, the target training stage, the target memory configuration parameter and the state information are respectively taken as graph nodes, and the association relationship between the plurality of graph nodes is taken as an edge, and a target graph is generated, including: defining the target training stage, the target memory configuration parameter and the state information as a first-level graph node, a second-level graph node and a third-level graph node respectively; based on the configuration association relationship between the target training stage and the target memory configuration parameter, and the state association relationship between the target memory configuration parameter and the state information, connecting the first-level graph node and the corresponding second-level graph node, and the second-level graph node and the corresponding third-level graph node, to obtain the target graph.

[0051] In the process of generating the target graph, the three core elements need to be clearly defined and the attributes need to be improved. Specifically, the target training stage obtained by analysis is defined as a first-level graph node, which is attached with associated attributes such as a stage unique identifier, a position of a training process, a corresponding parameter adjustment category, and a historical execution record, clearly outlining the training link characteristics of the fault occurrence.

[0052] The target memory configuration parameter is defined as a second-level graph node, which is supplemented with attributes such as a data type of the parameter, a value interval supported by hardware, a default configuration value, and a dependency relationship with other parameters, fully presenting the parameter itself information that causes the fault.

[0053] The state information is defined as a third-level graph node, which can include attributes such as a specific fault type of the training failure, a timestamp of the fault occurrence, a fault impact range, a corresponding fault code, and a detailed failure description text, accurately recording the state details of the training failure.

[0054] After the definition of the nodes is completed, the edges of the graph are constructed based on the internal logical relationship between the elements. First, for the first-level graph node and the second-level graph node, a connection edge is established according to the configuration association relationship therebetween. This configuration association relationship is essentially a corresponding relationship between the core function of the target training stage and the adjustment requirement of the target memory configuration parameter, that is, the core task of the target training stage is to adjust the target memory configuration parameter. Therefore, this connection edge will clearly mark the association type of "stage adjustment-target parameter", directly reflecting the functional corresponding logic of the first-level graph node and the second-level graph node.

[0055] For the second-level graph node and the third-level graph node, a connection edge is established based on the state association relationship therebetween. This state association relationship is a causal association, that is, the abnormal parameter value setting of the target memory configuration parameter directly leads to the training failure state described by the third-level graph node. Therefore, this connection edge will mark the association type of "parameter abnormality-triggered fault", clearly defining the causal relationship between the second-level graph node and the third-level graph node.

[0056] Through the construction mode of the node hierarchical definition and the directional association edge, a target graph with clear structure hierarchy and clear association logic is finally formed, thereby providing structured basic data support for subsequent graph training and adjustment strategy generation.

[0057] According to an embodiment of the present application, the target graph is trained to obtain an adjustment strategy for a parameter value of a target memory configuration parameter, including: performing feature extraction on graph information in the target graph to obtain a target feature vector representing characteristics of the target graph; and inputting the target feature vector into a decision module to process the target feature vector through the decision module to obtain the adjustment strategy.

[0058] In the process of training the target graph to obtain the adjustment strategy, feature extraction needs to be performed on the target graph to fully mine node attributes, association logic and structural characteristics in the target graph. In the extraction process, the core attributes of the nodes of each hierarchical graph are quantitatively processed.

[0059] For example, the stage number and adjustment parameter category of the first-level graph node (target training stage) are converted into numerical features. The value range, dependency relationship and current configuration value of the second-level graph node (target memory configuration parameter) are encoded. The fault code, influence range and failure description of the third-level graph node (state information) are converted into calculable feature vectors through semantic analysis.

[0060] At the same time, the association characteristics of the graph edges are captured, including the type identification of the edges (such as "stage adjustment-target parameter" and "parameter exception-triggered fault"), the association strength (based on the frequency weighting of the occurrence of the same type of association in historical cases), and the path characteristics between nodes (such as the connection path length from the first level to the second level, the influence weight of the indirect association nodes, etc.).

[0061] In addition, the overall structural characteristics of the graph are also extracted, such as the degree distribution of the graph nodes, the density of the associated edges, and the clustering coefficient of the core graph nodes. Through a feature fusion algorithm, the dispersed single-dimensional features are integrated into a unified target feature vector. The target feature vector completely represents the node attributes, association logic and structural form of the target graph, ensuring that the subsequent analysis can cover all key information related to the fault.

[0062] After the feature extraction is completed, the generated target feature vector is input into a preset decision module. The decision module has an analysis model trained based on a large number of historical memory training fault cases, and is associated with a mapping database of historical fault graph feature vectors and corresponding successful adjustment strategies.

[0063] The decision module performs similarity matching on the input target feature vector, retrieves cases highly consistent with the current feature vector from the historical database, and extracts effective adjustment experience in these cases. Subsequently, the graph features of the current target graph are combined for targeted optimization, such as considering the real-time hardware adaptation of the current target memory configuration parameters, the execution environment differences of the target training stage, the special manifestations of faults in the state information, etc., to fine-tune the adjustment direction and amplitude in the historical experience.

[0064] At the same time, the decision module also checks the feasibility of the adjustment strategy, such as whether the adjusted parameter value is within the value range supported by the hardware, whether it will cause configuration conflicts of other associated parameters, and whether it meets the parameter adjustment constraints of the target training stage.

[0065] After multiple rounds of matching, optimization and verification, the decision module finally outputs the adjustment strategy, which explicitly includes the parameter adjustment type, parameter adjustment amount, adjustment order, and adjusted verification index of the target memory configuration parameters, as well as the backup adjustment scheme if the first adjustment is unsuccessful, ensuring that the adjustment strategy can effectively solve the current memory training failure problem.

[0066] By extracting features from the target graph and then inputting the target feature vector into the decision module for comprehensive analysis and evaluation, the decision efficiency and accuracy of the adjustment strategy are effectively improved.

[0067] According to the embodiments of the present application, the memory training method further comprises: in response to receiving the update training log representing the memory training failure based on the updated parameter value transmitted by the basic input and output system, generating a feedback signal representing the invalidation of the adjustment strategy based on the update training log; based on the feedback signal, determining the target graph and the adjustment strategy as negative training samples; and updating the parameters of the decision module using the negative training samples.

[0068] The management controller continuously monitors the training result feedback of the basic input and output system. When receiving the update training log representing the failure of the memory training based on the updated parameter value, it first checks whether the target training stage, target memory configuration parameter and previous processed fault object recorded in the update training log are consistent to confirm that this failure is caused by the invalidation of the previous adjustment strategy, rather than a new training failure or parameter transmission abnormality.

[0069] Based on the updated state information in the training log that characterizes the failure (such as the parameter adjustment still exceeding the hardware compatibility range, the fault type changing from parameter value out of bounds to timing mismatch, etc.), combined with the previously generated target graph identification and adjustment strategy, a feedback signal is generated to characterize the invalidity of this round of adjustment strategy. This feedback signal will be accompanied by key metadata such as fault association link, adjustment strategy execution result, and new failure state summary, ensuring that the subsequent processing can accurately locate the complete failure scenario corresponding to the invalid strategy.

[0070] After generating the feedback signal, the management controller will integrate the core data in this failure handling process into negative training samples, including the original target graph (containing graph node attributes and associated edge information at each level), the adjustment strategy output in the last round (including parameter adjustment direction, amplitude, and verification standard), and the key failure information extracted from the updated training log (new state information, actual execution effect of updated parameters, and specific performance of un-repaired faults).

[0071] At the same time, the negative training sample is labeled with "strategy invalid" and associated with the corresponding failure scenario features (such as hardware model, current running environment parameters, and training process version), ensuring that the sample has clear invalidation labeling and complete context information, providing sufficient basis for parameter optimization of the decision module.

[0072] Subsequently, the management controller inputs the constructed negative training sample into the model update channel of the decision module. The decision module will first perform feature alignment processing on the negative training sample, matching the target graph features and adjustment strategy features in the sample with the feature dimensions inside the model, and then combining the invalidation labeling in the negative sample to analyze the mismatch points between the adjustment strategy and the target graph features.

[0073] For example, it may be that the parameter adjustment direction is incorrect, the adjustment amplitude is insufficient, or the interaction between the target memory configuration parameter and other associated parameters is not considered, or the deep failure reason implied by the failure state in the update training log is ignored.

[0074] Based on these analysis results, the decision module will start the built-in model parameter iteration process, adjust the weights of the corresponding feature vectors and adjustment strategy mapping relationship in the model through the backpropagation algorithm, reduce the probability of outputting the invalid strategy when encountering similar graph features in the future, and at the same time, strengthen the learning of effective strategy direction.

[0075] During the parameter update process, cross-validation will also be performed with the historical accumulation of positive and negative training sample library to ensure that the updated model optimizes the invalid strategy judgment while not affecting the processing capability of other effective failure scenarios, finally realizing the self-adaptive iteration of the decision module and improving the output accuracy of the adjustment strategy for similar memory training failures in the future.

[0076] According to an embodiment of the present application, the memory training method further comprises: in the case where the plurality of target training stages exist, determining a root cause stage causing the failure from the plurality of target training stages by analyzing the time sequence relationship between the plurality of target training stages; generating a root cause graph by taking the root cause stage, the root cause memory configuration parameter corresponding to the root cause stage, and the state information associated with the root cause memory configuration parameter as graph nodes respectively, and taking the association relationship between the plurality of graph nodes as edges; and training the root cause graph to obtain an adjustment strategy for the parameter value of the root cause memory configuration parameter.

[0077] When the plurality of target training stages exist, the time sequence relationship of all target training stages is analyzed, the execution time stamp of each target training stage, the dependency identification between stages (such as the precondition that the parameter of the previous stage takes effect is the precondition for the start of the next stage), and the training result feedback of each stage (including partial failure, complete failure, and failure impact range) are extracted in combination with the preset execution process of memory training, and a complete time sequence link graph is constructed.

[0078] Through time sequence link tracking, the propagation path of the fault is investigated. If the training of a stage fails, the subsequent associated stages all appear a chain failure caused by abnormal parameter configuration of the stage, or the failure state of the stage cannot be indirectly repaired through parameter adjustment of the subsequent stages, then the stage is determined as a root cause stage causing the overall training failure. At the same time, the secondary fault stage (i.e. only itself fails but does not affect other stages, or a failure stage derived from the root cause stage) is excluded to ensure the accuracy of positioning of the root cause stage.

[0079] After the root cause stage is determined, the root cause memory configuration parameter corresponding to the root cause stage is determined, that is, the memory configuration parameter that the root cause stage is mainly responsible for adjusting. At the same time, the state information associated with the root cause memory configuration parameter is integrated. Then, the root cause graph is generated: taking the root cause stage, the root cause memory configuration parameter, and the associated state information as core graph nodes respectively, constructing the edges of the graph based on the internal logical association between the nodes, and finally forming a root cause graph with complete structure and clear causal relationship.

[0080] The generated root cause graph is trained, and finally an adjustment strategy for the root cause memory configuration parameter is output to ensure that the memory training failure problem can be solved from the root after adjustment. In the multi-stage failure scenario, the root cause stage is located through time sequence analysis, and a root cause graph with stronger pertinence is constructed, which effectively improves the diagnosis accuracy of complex faults and reduces the waste of computing resources.

[0081] According to an embodiment of the present application, the adjustment strategy includes a parameter adjustment type and a parameter adjustment amount; and according to the adjustment strategy, updating a parameter value stored in the shared memory by the target memory configuration parameter includes: obtaining a current parameter value of the target memory configuration parameter from the shared memory; determining an updated parameter value after adjustment of the current parameter value according to the parameter adjustment type and the parameter adjustment amount, wherein the parameter adjustment type is increasing or decreasing; and comparing the updated parameter value with a preset parameter threshold corresponding to the target memory configuration parameter; and in the case that the updated parameter value is within the preset parameter threshold, updating the parameter value of the target memory configuration parameter stored in the shared memory to the updated parameter value.

[0082] In the process of updating the target memory configuration parameter in the shared memory according to the adjustment strategy, it is necessary to clearly define the parameter adjustment type (specifically, increasing or decreasing) and the corresponding parameter adjustment amount. The parameter adjustment amount can be in the form of a specific numerical value, a percentage or a step increment, and has been accurately quantified in combination with the properties (such as integer type, floating point type) of the target memory configuration parameter and the hardware adaptation requirements.

[0083] The management controller will locate the storage address of the target memory configuration parameter in the shared memory through a preset shared memory access interface, and the storage address is determined by the management controller and the basic input / output system in advance, and has a unique identification and access permission isolation mechanism. When obtaining the current parameter value, the management controller will first perform data integrity check on the storage address, and load it to the local cache after confirming that it is correct, as the basis data for subsequent calculation.

[0084] Subsequently, according to the parameter adjustment type and the parameter adjustment amount specified in the adjustment strategy, the updated parameter value is calculated. If the parameter adjustment type is increasing, the current parameter value in the local cache is added to the parameter adjustment amount. If the parameter adjustment amount is in the form of a percentage, the current parameter value is multiplied by the percentage coefficient to obtain an increment value, and then summed with the current parameter value. If the parameter adjustment type is decreasing, subtraction is performed, and the percentage form of the parameter adjustment amount is processed in the same way. The calculation process will strictly follow the data type specification of the target memory configuration parameter to ensure the accuracy of the calculation result and avoid invalid parameters due to data overflow or precision deviation.

[0085] After the updated parameter value is calculated, it needs to be compared with the preset parameter threshold corresponding to the target memory configuration parameter, which is determined based on the memory specification provided by the hardware manufacturer, long-term stability test data and actual application scenario requirements. Specifically, it includes the minimum effective threshold and the maximum safe threshold, forming a clear effective value interval.

[0086] In the comparison process, it is checked whether the updated parameter value is greater than or equal to the minimum effective threshold value and less than or equal to the maximum safe threshold value, to ensure that the parameter after adjustment meets the basic requirements of hardware operation and does not exceed the safe operation range to cause hardware damage or performance abnormalities.

[0087] When it is confirmed that the updated parameter value is within the preset parameter threshold value, the management controller sends an exclusive access request to the shared memory, to avoid encountering concurrent reading of the basic input / output system or writing operation of other modules during the updating process, and to ensure the atomicity of the updating operation. Then, the calculated updated parameter value is written into the storage address to overwrite the original parameter value, and the corresponding check information of the address is updated synchronously.

[0088] After the updating is completed, the parameter value of the address is read again for secondary checking, to confirm that the written data is completely consistent with the updated parameter value and there is no transmission or storage error. Finally, the updating of the target memory configuration parameter stored in the shared memory is completed, to provide accurate and effective parameter support for the subsequent re-execution of the memory training of the basic input / output system.

[0089] According to the embodiments of the present application, according to the adjustment strategy, the updating of the parameter value stored in the shared memory of the memory configuration parameter further includes: in the case that the updated parameter value is outside the preset parameter threshold value, updating the parameter value of the target memory configuration parameter stored in the shared memory to the boundary parameter value of the preset parameter threshold value closest to the updated parameter value.

[0090] When the updated parameter value is outside the preset parameter threshold value after the comparison and confirmation, the management controller first determines the specific direction of the parameter overflow, by performing secondary accurate comparison of the updated parameter value with the minimum effective threshold value and the maximum safe threshold value of the preset parameter threshold value, to determine whether the overflow is below the lower limit of the minimum effective threshold value or above the upper limit of the maximum safe threshold value. The two boundary values of the preset parameter threshold value are determined based on the physical characteristics of the memory hardware, the safe operation specification provided by the manufacturer, and the long-term stability test data, and are the critical effective values for guaranteeing that the memory hardware is not damaged and can maintain the basic operation ability, and have clear hardware adaptability and safety.

[0091] After the overflow direction is determined, the management controller determines the boundary parameter value closest to the updated parameter value. If it is determined that the lower limit is exceeded (i.e., the updated parameter value is less than the minimum effective threshold value), the boundary parameter value closest to the updated parameter value is the preset minimum effective threshold value. If it is determined that the upper limit is exceeded (i.e., the updated parameter value is greater than the maximum safe threshold value), the boundary parameter value closest to the updated parameter value is the preset maximum safe threshold value. By selecting the boundary parameter value, a balance is sought between the parameter adjustment requirement and the hardware safety constraint, to maximize the adjustment intention of the adjustment strategy and to avoid memory hardware failure, signal disorder, or complete interruption of the training process due to parameter exceeding the safe range.

[0092] According to an embodiment of the present application, the memory training method further comprises: recording the number of repetitions of the memory training for the same target memory configuration parameter by a counter; and in the case that the number of repetitions reaches a preset number threshold and the memory training still fails, suspending the adjustment of the parameter value of the target memory configuration parameter, encapsulating the corresponding historical target map, historical adjustment strategy and historical training log as an abnormal data packet, and uploading the abnormal data packet to the operation and maintenance management platform.

[0093] The management controller maintains a counter for each target memory configuration parameter, which is initialized and cleared when the memory training corresponding to the target memory configuration parameter is first detected to fail and the adjustment process is started. The counter is automatically incremented each time the memory training is performed based on the updated parameter value and is determined to fail. The counter is strictly associated with the unique identifier of the target memory configuration parameter during the counting process to avoid confusion with the training number of other parameters.

[0094] Meanwhile, the counter has a dynamic reset mechanism. If the memory training corresponding to the updated parameter succeeds during the repeated training process, or the target memory configuration parameter is changed due to a fault scenario switch, the counter is immediately reset to zero to ensure that the counting result only reflects the number of consecutive training failures for the same parameter.

[0095] The preset number threshold is not a fixed value, but is determined comprehensively in combination with the stability limit of the memory hardware, the average number of repairs of historical similar faults, the response cycle of operation and maintenance investigation, and other factors. It avoids prematurely terminating effective adjustment attempts due to a too low threshold, and prevents invalid training cycles from occupying system resources due to a too high threshold.

[0096] The management controller monitors the cumulative value of the counter in real time. After each failed memory training is completed, the current number of repetitions is compared with the preset number threshold, and it is confirmed that the latest training log still represents training failure. Only when both conditions are met, the subsequent abnormal handling process is triggered.

[0097] Upon determining that the number of repetitions reaches the preset threshold and the training still fails, the management controller immediately suspends all adjustment operations for the target memory configuration parameter, including stopping the generation of new adjustment strategies, prohibiting any updates to the parameter values in the shared memory, to avoid invalid parameter adjustments continuously consuming hardware resources or causing more complex system abnormalities.

[0098] Subsequently, the full amount of historical data related to the target memory configuration parameter is collected, wherein the historical target map covers the original map generated after each training failure, the root cause map (if there is a multi-stage failure scenario). The historical adjustment strategy includes the parameter adjustment type, the parameter adjustment amount, the effective parameter value, the strategy generation timestamp and other details output in each round. The historical training log contains complete records of the execution timing, failure state information, fault code, hardware environment data (such as temperature, load) and other complete records of each training.

[0099] In the packaging process, necessary metadata is added to the abnormal data packet, including device unique identification, detailed information of target memory configuration parameter (name, ID, hardware associated module), fault starting time, cumulative training failure duration, preset number threshold and the like. At the same time, integrity check processing is performed on the abnormal data packet, and a special check code is generated to ensure that the data is not tampered with or lost during transmission.

[0100] Finally, the management controller uploads the packaged abnormal data packet to the platform according to the communication protocol stipulated by the operation and maintenance management platform through a preset secure communication link (supporting data encryption transmission). During the uploading process, the transmission state is monitored in real time, and if the transmission is interrupted or fails, a retransmission mechanism is automatically triggered until the platform feedback is received successfully, so that the operation and maintenance personnel can accurately locate the fault root cause based on the complete historical data in the data packet, and develop targeted troubleshooting and repair solutions.

[0101] According to the embodiment of the application, before the target map is generated by taking the target training stage, the target memory configuration parameter and the state information as the graph nodes and taking the association relationship between the plurality of graph nodes as the edge, the method further comprises: matching in the local adjustment knowledge base taking the target training stage, the target memory configuration parameter and the state information as the query condition to generate a matching result, wherein the local adjustment knowledge base stores a successful adjustment record, and the successful adjustment record includes a mapping relationship between the training stage and the memory configuration parameter corresponding to the training stage; and in a case where the matching result represents that there is a corresponding successful adjustment record, writing the parameter value in the successful adjustment record into the shared memory.

[0102] Before the target map generation process is performed, the management controller will first start the matching query process of the local adjustment knowledge base, which includes complete solutions of various training stages, corresponding memory configuration parameters and adapted fault states under different training scenarios. Each successful adjustment record not only contains the mapping relationship between the training stage and the corresponding memory configuration parameter, but also carries key metadata such as the effective environment condition (such as hardware model, running temperature, load condition) of the parameter value, the stable running duration after adjustment, the cooperative configuration information of the associated parameters and the like, to ensure the accuracy and practicability of the query matching.

[0103] In the query process, the management controller takes the target training stage, target memory configuration parameters and state information obtained by analysis as the core query condition, and combines real-time data of the current hardware environment (such as the model identification of the current memory module, the system running temperature, and the current hardware load rate) to supplement the query dimension, thereby forming a multi-dimensional composite query condition.

[0104] Through the built-in efficient retrieval algorithm, the local adjustment knowledge base is traversed, the target training stage and the target memory configuration parameters are used for preliminary screening, and the candidate successful adjustment records corresponding to the same type of training stage and the same parameters are locked. Then, the fault type and fault performance in the state information are precisely matched to determine whether the fault scene in the candidate record is highly consistent with the current training failure state information. Finally, an explicit matching result is generated, which indicates whether there is a completely matched successful adjustment record, a matching degree score, and a corresponding record identifier.

[0105] When the matching result indicates that there is a corresponding successful adjustment record, the management controller extracts the verified effective parameter value from the successful adjustment record, and retrieves the associated configuration information in the record to determine whether the parameter value needs to be adjusted in coordination with other associated memory configuration parameters.

[0106] The extracted effective parameter value is written into the shared memory to overwrite the original parameter value, thereby completing the parameter update operation of the shared memory. Thereafter, the subsequent target graph generation and training process is no longer needed, and the basic input and output system is directly provided with the effective parameter value verified by practice, so that a new round of memory training is quickly started, thereby greatly improving the fault repair efficiency and the training success rate.

[0107] According to the embodiments of the present application, the memory training method further comprises: in the case where it is determined that the memory training based on the updated parameter value is successful, establishing a mapping relationship between the updated parameter value stored in the current shared memory and the target training stage corresponding to the adjustment strategy triggering the current successful training and the updated parameter value of the target memory configuration parameter, generating a corresponding successful adjustment record, and storing it to the local adjustment knowledge base.

[0108] When the management controller receives the feedback signal transmitted by the basic input and output system, which indicates that the memory training based on the updated parameter value is successful, it checks whether the target training stage identifier and the target memory configuration parameter in the feedback signal are consistent with the object corresponding to the current adjustment operation. At the same time, the state data after the training is successful (such as the memory read-write speed, signal stability, and fault-free running time) is retrieved to confirm that this success is the effective training result achieved by the target memory configuration parameter after adjustment, rather than an indirect success caused by accidental triggering or other parameter adjustment, thereby ensuring the accuracy of the record.

[0109] The management controller accurately reads the current stored updated parameter value from the shared memory, and integrity check is performed on the data during the reading process. The target training phase corresponding to the current successful training, the target memory configuration parameter, and the updated parameter value confirmed after the check are directionally associated to form a core mapping link of "target training phase-target memory configuration parameter-updated effective parameter value". At the same time, the key context information of this adjustment is supplemented, including the original failure state information triggering this adjustment, the parameter adjustment type and adjustment amount of the adjustment strategy, and the verification index data of the successful training.

[0110] Based on the above mapping relationship and supplementary information, the management controller generates a standardized successful adjustment record. In addition to the core mapping content, the record also adds metadata such as unique identifier, record generation timestamp, basic input / output system version number, management controller firmware version, etc., to ensure that each record is traceable and unique.

[0111] After generating the successful adjustment record, the built-in deduplication algorithm is used to compare the existing successful adjustment records in the local knowledge base to avoid redundant storage of records with the same scene and the same parameter configuration.

[0112] The management controller writes the successful adjustment record that passes the check and is deduplicated into the local adjustment knowledge base, and establishes an index during the writing process to facilitate fast matching during subsequent queries. At the same time, the index directory of the knowledge base is updated to ensure that new records can be efficiently retrieved, and the written data is backed up to prevent record loss due to local storage failure. Finally, the storage of the successful adjustment record is completed, providing more abundant and accurate historical experience support for the rapid matching and solving of similar memory training failures in the future.

[0113] According to the embodiments of the present application, the training log is parsed, including: determining a log line including a preset error identifier from the training log; extracting a target memory configuration parameter associated with the preset error identifier from the log line, and reading a numerical field and a state description field corresponding to the target memory configuration parameter; performing semantic recognition on the state description field to determine state information representing failure of parameter value setting of the target memory configuration parameter; and determining a training failure field based on the target memory configuration parameter and the state information.

[0114] The preset error identifier is uniformly defined based on common fault types of memory training, log output specifications of the basic input / output system in advance, and may be a specific fault code, a key field combination, or a dedicated identifier string (such as "FAILED"), which has unique identification and can accurately lock the log content related to training failure.

[0115] The management controller scans the training log row by row in chronological order of log generation time, compares whether each log row contains a preset error identifier through a built-in string matching algorithm, and filters out irrelevant log rows for recording normal training progress, system state notification, etc., to ensure that only target log rows directly related to training failure are selected, avoiding irrelevant information interfering with the analysis result.

[0116] After finding the log row containing the preset error identifier, the target memory configuration parameter associated with the preset error identifier is extracted based on the field format specification of the log (which is determined by the management controller and the basic input / output system in advance, and clearly specifies the delimiters, position indexes, and keyword identifiers of each field).

[0117] Specifically, the log row is divided into different functional fields by a specific delimiter (such as a comma or a colon), and the field associated with the error identifier or through a keyword (such as "PARAM_NAME") is the target memory configuration parameter. When extracting, the name, unique identifier, and basic information of the target memory configuration parameter such as the parameter category are recorded.

[0118] At the same time, the numerical field and the state description field corresponding directly to the target memory configuration parameter are read synchronously. The numerical field usually contains the current configuration value of the target memory configuration parameter, the preset value range supported by the hardware, the adjustment record in the current training process, etc. The state description field is presented in natural language text form, which records the abnormal performance in the parameter configuration process in detail (such as "parameter value exceeds maximum threshold", "signal does not respond after configuration", "timing conflict with associated parameters", etc.).

[0119] The state description field is subjected to semantic recognition processing, and the state description field is converted into standardized and structured state information, which clearly represents the core reason, specific performance, and associated impact of the failure of the parameter value setting of the target memory configuration parameter. For semantic processing, the management controller calls a built-in lightweight natural language processing model that has been pre-trained and optimized through a large number of memory training fault description texts, which can accurately identify the core semantic information in the state description.

[0120] Based on the extracted target memory configuration parameter and the state information obtained through semantic recognition, field integration and standardization processing are performed to finally form a training failure field with unified structure and comprehensive information, providing accurate and structured basic data support for subsequent target training stage positioning, target graph generation, etc.

[0121] According to an embodiment of the present application, the memory training method further comprises: after completing the updating of the parameter value stored in the shared memory, the basic input / output system sends a parameter ready signal, so that the basic input / output system, in response to receiving the parameter ready signal, restarts and re-reads the updated parameter value from the shared memory, and re-executes the memory training based on the updated parameter value.

[0122] When the management controller completes the updating of the target memory configuration parameter value in the shared memory, the management controller sends a parameter ready signal through a communication link previously agreed with the basic input / output system. The basic input / output system continuously monitors the signal transmission of the communication link, and when receiving the parameter ready signal, the basic input / output system starts to restart.

[0123] After the restart is completed, the basic input / output system re-reads the updated parameter value from the shared memory, loads the parameter value into the register of the memory training module, synchronously configures the hardware logic associated with the parameter, and ensures that the hardware level is adapted to the new parameter value. Subsequently, the basic input / output system executes the memory training based on the updated parameter value, and generates a detailed training log, so as to feed back the final result to the management controller after the training is completed, and form a complete closed loop of parameter adjustment and training verification.

[0124] Figure 3 A flowchart of a memory training method applied to a management controller according to another embodiment of the present application is shown.

[0125] As shown in Figure 3 , this embodiment comprises operations S301 to S311.

[0126] In operation S301, a training log characterizing a failure of memory training transmitted by the basic input / output system is received.

[0127] In operation S302, the training log is parsed to determine the target training stage in which the failure occurs, the target memory configuration parameter, and the state information characterizing the failure of setting the parameter value of the target memory configuration parameter.

[0128] In operation S303, it is determined whether the target training stage includes multiple target training stages. If yes, operation S304 is executed; if no, operation S305 is executed.

[0129] In operation S304, the timing relationship between the multiple target training stages is analyzed to determine the target training stage causing the failure from the multiple target training stages.

[0130] In operation S305, a target graph is generated by taking the target training stage, the target memory configuration parameter, and the state information as graph nodes, and taking the association relationship between the multiple graph nodes as edges.

[0131] At operation S306, the target graph is trained by using the decision module to obtain an adjustment strategy of the parameter value of the target memory configuration parameter.

[0132] At operation S307, according to the adjustment strategy, the parameter value of the memory configuration parameter stored in the shared memory is updated.

[0133] At operation S308, it is judged whether an update training log representing that the memory training based on the updated parameter value fails is received by the basic input and output system. If yes, operation S309 is performed; if no, the process ends.

[0134] At operation S309, based on the update training log, a feedback signal representing that the adjustment strategy is invalid is generated.

[0135] At operation S310, based on the feedback signal, the target graph and the adjustment strategy are determined as negative training samples.

[0136] At operation S311, the parameters of the decision module are updated by using the negative training samples.

[0137] According to the embodiments of the present application, by real-time monitoring of the transmission data of the basic input and output system, when the training log representing that the memory training fails is received, the fault related data is accurately locked by the log capture mechanism, ensuring the integrity and real-time of the log information, and providing reliable data source for subsequent analysis.

[0138] The obtained training log is analyzed and processed, the key information is extracted by means of the log analysis algorithm, the target training stage where the failure occurs is accurately located, the corresponding target memory configuration parameter is identified, and the state information representing the failure of the parameter value setting is extracted. In the analysis process, invalid data is filtered by the data verification mechanism to ensure the accuracy of the extracted information.

[0139] Then the number of target training stages is detected. If it is determined that there are multiple target training stages, the time sequence analysis process is triggered. By means of the time sequence track tracking technology, the execution order and the dependency relationship between stages are sorted out, and the root stage causing the memory training failure is accurately located from multiple stages by combining the fault conduction path analysis.

[0140] If there is only a single target training stage, the graph construction link is directly entered. In the graph generation process, the target training stage, the target memory configuration parameter and the state information are respectively mapped as independent graph nodes. The internal logical association between nodes is identified and extracted by means of the association relationship mining algorithm, the association relationship is converted into the edge of the graph, and the structured target graph is constructed to clearly present the association mode of the fault related elements.

[0141] The target graph is trained by using a preset decision module. The decision module is built-in with a deep learning model. By performing feature learning and pattern recognition on the fault information carried by the nodes and edges in the graph, a parameter value adjustment strategy for the target memory configuration parameters is automatically generated. The adjustment strategy is targeted and executable, and can accurately point to the parameter configuration problem.

[0142] After the parameter value is updated, the transmission data of the basic input / output system is continuously monitored to determine whether an update training log indicating that the memory training based on the updated parameter value still fails is received. If the update training log is not received, it indicates that the adjustment strategy is effective, and the process ends.

[0143] If the update training log is received, a feedback signal generation mechanism is triggered to generate a feedback signal indicating that the current adjustment strategy is invalid. The signal carries key identification information of the unresolved fault. At the same time, the target graph and the corresponding adjustment strategy are marked as negative training samples. Through a negative sample reinforcement learning mechanism, the model parameters of the decision module are iteratively updated using the negative training samples.

[0144] Through two parameter update operations (triggered based on the feedback signal and directly driven based on the negative training sample), the learning ability of the decision module is further optimized, the decision bias of the model is corrected, and the accuracy and effectiveness of the subsequently generated adjustment strategy are improved, realizing self-improvement and performance improvement of the decision module.

[0145] Figure 4 A flowchart of a memory training method applied to a basic input / output system according to an embodiment of the present application is shown.

[0146] As shown in Figure 4 This embodiment includes operation S410 to operation S450.

[0147] In operation S410, the parameter value of the memory configuration parameter is read from the shared memory, wherein the shared memory is accessed by the basic input / output system and the management controller.

[0148] In operation S420, the memory training is performed based on the parameter value.

[0149] In operation S430, in the case of failure of the memory training, a training log indicating the failure of the memory training is transmitted to the management controller.

[0150] In operation S440, in response to the management controller completing the update of the parameter value in the shared memory, the updated parameter value is re-read from the shared memory.

[0151] In operation S450, the memory training is re-performed based on the updated parameter value.

[0152] According to an embodiment of the present application, the basic input / output system obtains target memory configuration parameters by reading the shared memory previously negotiated with the management controller, and loads the corresponding parameter values into the dedicated registers of the memory training module to provide basic configuration data for subsequent training.

[0153] When performing memory training based on the read parameter values, the basic input / output system starts operation according to the preset training process, first configures the parameter values to the memory controller, the timing generator and other related hardware modules, and then sequentially performs multiple training stages such as timing calibration, signal gain adjustment, voltage adaptation verification, etc. Each stage will monitor the running state data fed back by the hardware in real time, such as signal integrity, timing synchronization accuracy, data read / write response speed, etc.

[0154] During the training process, the actual monitoring data is continuously compared with the preset training qualified threshold to determine whether the training of each stage achieves the expected effect, while the execution progress, parameter taking effect and key monitoring indicators of each stage are recorded to form a complete training process record.

[0155] If the actual data exceeds the qualified threshold or the hardware module feeds back abnormal response in a certain training stage, it is determined that the memory training fails, and the basic input / output system will immediately terminate the current training process and generate a training log by integrating the key information recorded during the training process.

[0156] The training log contains core contents such as preset error identifier, specific stage number of training failure, currently used memory configuration parameter value, hardware state data at the time of failure, failure performance description, etc. After being packaged in a unified format specification, it is transmitted to the management controller through the communication link agreed with the management controller to ensure that the log information is complete and accurate, and to provide a comprehensive basis for the management controller to analyze the failure subsequently.

[0157] The basic input / output system will continuously monitor the parameter update feedback signal of the management controller, and when receiving the parameter update completion notification sent by the management controller, the basic input / output system will start the shared memory reading process again to read the updated parameter values from the shared memory. After obtaining the updated parameter values, the basic input / output system will reset the state of the memory training module, clear the failure records and intermediate data of the previous training, and then reconfigure the updated parameter values to the related hardware modules.

[0158] At this time, targeted training will be started for the training stage that failed in the last round, if the training of this stage is successful, the subsequent unfinished training process will be continued. If there is still an abnormality, a new training state will be recorded according to the same logic. During the entire retraining process, the parameter taking effect and the monitoring data of each stage will be recorded in real time to form a new training record, so as to provide the management controller with an updated training log when the training fails again, and to support the optimization of subsequent adjustment strategies.

[0159] By reading the memory configuration parameter value from the shared memory, efficient cooperation between the basic input / output system and the management controller is realized. When the memory training fails, the training log is timely transmitted to the management controller, facilitating the management controller to quickly locate the failure problem. After the management controller updates the parameter value, the training is re-read and executed, effectively improving the success rate of memory training and fault repair efficiency, and ensuring the stable operation of the system.

[0160] According to the embodiment of the application, before performing the operation of reading the parameter value of the memory configuration parameter from the shared memory, the method further comprises: reading a memory training result flag bit stored in the non-volatile memory; in the case that the memory training result flag bit is a first identification, performing the operation of reading the parameter value of the memory configuration parameter from the shared memory; in the case that the memory training result flag bit is a second identification, reading a historical successful parameter value from the non-volatile memory, and performing memory training based on the historical successful parameter value.

[0161] Before performing the operation of reading the parameter value of the memory configuration parameter from the shared memory, the basic input / output system will first start the access process of the non-volatile memory, which is a special storage medium pre-bound with the basic input / output system, for persistently storing key configuration information related to memory training.

[0162] The basic input / output system accesses the non-volatile memory through a preset hardware interface and communication protocol, and accurately locates the storage address corresponding to the memory training result flag bit. The memory training result flag bit is a standardized defined state identifier, which only occupies a fixed byte length (such as 1 byte), and the first identification and the second identification are jointly agreed by the basic input / output system and the management controller (such as the first identification is 0 and the second identification is 1), which correspond to different memory training history states.

[0163] When reading the flag bit data, the specific type of the flag bit is determined. If it is determined to be the first identification, it means that there is no successful memory training before (such as the first time the device is started, the last memory training fails, the hardware configuration changes, etc.), at this time the basic input / output system will perform the operation of reading the parameter value of the memory configuration parameter from the shared memory according to the preset process, to obtain the initial configuration parameter or the updated adjustment parameter provided by the management controller, to provide basic data for the first training or retry training.

[0164] If it is determined that the memory training result flag bit is the second identification, it means that there is a successful memory training before, and the corresponding effective parameter is stored in the non-volatile memory. At this time, the basic input / output system will locate the storage area of the historical successful parameter value in the non-volatile memory based on the preset address mapping relationship.

[0165] The basic input / output system directly loads the historical success parameter value from the storage area to a dedicated register of the memory training module, without accessing the shared memory to obtain the parameter again, and then starts a memory training process based on the parameter value, thereby greatly shortening the training start time, improving the training success rate and system start efficiency by reusing the verified effective parameter.

[0166] According to an embodiment of the present application, after performing the re-executed memory training, the method further comprises: in the case of determining that the memory training is successful, writing the parameter value used when the current training is successful into the non-volatile memory, and updating the memory training result flag bit to a second identification.

[0167] After the basic input / output system re-executes the memory training based on the updated parameter value, it continuously monitors the execution state of each core stage of the memory training, including the synchronization accuracy of timing calibration, the integrity of signal transmission, the response speed and error-free rate of data reading and writing, and other key indicators, to ensure that all indicators are stably within the preset qualified threshold range.

[0168] After determining that the training is successful, the basic input / output system reads the updated parameter value used in the current training from the shared memory again, and collects the associated context information corresponding to the updated parameter value, including the target training stage corresponding to the current training, the timestamp of successful training, the current hardware environment data, and the like. These information will be stored together with the updated parameter value to provide adaptability reference for subsequent reuse of the parameter.

[0169] The basic input / output system encapsulates the current successful updated parameter value and the associated context information into a data block according to a preset format, writes it into a specified storage area through a standardized communication protocol, and synchronously generates a check code of the data block stored in an adjacent address for integrity verification when read subsequently.

[0170] After writing is completed, the basic input / output system locates the storage address of the memory training result flag bit in the non-volatile memory, and updates the flag bit that may originally be the first identification to a preset second identification. The second identification is a state identification agreed upon in advance with the management controller, representing the existence of an effective historical success parameter. The writing result of the flag bit is checked again during the updating process to ensure accurate switching of the identification.

[0171] Through the updating of the identification, an explicit guide is provided for subsequent system start or memory training, so that the initial parameter does not need to be repeatedly obtained from the shared memory or the complex training process needs to be performed during the next start, and the historical success parameter in the non-volatile memory can be directly reused, thereby greatly improving the system start efficiency and the stability of the memory training.

[0172] According to an embodiment of the present application, a system restart operation is performed in response to receiving a parameter ready signal sent by the management controller; and updated parameter values are read from the shared memory to re-perform memory training in case of completion of the restart.

[0173] The basic input / output system continuously monitors signal transmission on the system management bus, and sends a parameter ready signal through the bus after the management controller completes updating of parameter values in the shared memory. The basic input / output system starts a targeted system restart operation after receiving the parameter ready signal.

[0174] During the restart process, the basic input / output system gradually resets register states of target modules, clears historical training cache data, and reloads firmware logic related to memory training, to ensure that each module is in an initial state to welcome a new round of parameter configuration.

[0175] After the restart is completed, a targeted memory training process is performed, which focuses on the training phase of the previous round of failure, and forms a detailed training process record, so as to feed back the training log to the management controller after the training is completed. This process effectively avoids training failure caused by untimely parameter update, improves the reliability of memory training and the stability of the whole system, reduces the time required for troubleshooting and repair, and improves the operation efficiency of the system.

[0176] Figure 5 A flowchart of interaction between the management controller and the basic input / output system according to an embodiment of the present application is shown.

[0177] As shown in Figure 5 , this embodiment includes operations S1-S8.

[0178] In operation S1, a memory training result flag bit is read.

[0179] In operation S2, if it is the first identification, parameter values of memory configuration parameters are read.

[0180] In operation S3, memory training is performed based on the parameter values, and a training log is generated.

[0181] In operation S4, if the training fails, the training log is transmitted.

[0182] In operation S5, the training log is analyzed to obtain an adjustment strategy for the parameter values of the target memory configuration parameters, and the updated parameter values are determined.

[0183] In operation S6, a parameter ready signal is sent.

[0184] In operation S7, the updated parameter values are read, and memory training is re-performed.

[0185] At operation S8, if the training is successful, the memory training result flag is updated to the second identification.

[0186] According to the embodiment of the present application, by switching the state of the flag, each stage of the training process is accurately controlled. The reading of the parameter value and the execution of the training ensure the pertinence of the memory training. The transmission and analysis of the training log realize the effective transmission of the failure or adjustment demand. The parameter readiness signal and the reading of the updated parameter value guarantee the timely training verification after the parameter adjustment. Finally, the update of the flag clearly marks the training state.

[0187] The overall process realizes the closed-loop management of the memory training from parameter reading, training execution, failure feedback to parameter update and retraining. Through the accurate cooperation of each link, the efficiency and accuracy of the memory training are effectively improved, and the optimization and stability of the memory configuration are ensured.

[0188] Based on the above memory training method, the present application further provides a memory training device applied to a management controller and a memory training device applied to a basic input / output system.

[0189] Figure 6 The structure block diagram of the memory training device applied to the management controller according to the embodiment of the present application is shown.

[0190] As shown in Figure 6 the memory training device 600 applied to the management controller of this embodiment includes a log analysis module 610, a graph construction module 620, a strategy generation module 630 and a parameter update module 640.

[0191] The log analysis module 610 is used to analyze the training log to determine the target training stage where the failure occurs and the training failure field describing the training failure state in response to receiving the training log characterizing the memory training failure transmitted by the basic input / output system, wherein the memory training passes through a plurality of training stages in sequence, the plurality of training stages adjust different types of memory configuration parameters respectively, and the training failure field includes a target memory configuration parameter and state information characterizing the failure of the parameter value setting of the target memory configuration parameter. In an embodiment, the log analysis module 610 can be used to perform the operation S210 described in the foregoing, and details are not repeated here.

[0192] The graph construction module 620 is used to generate a target graph by taking the target training stage, the target memory configuration parameter and the state information as graph nodes respectively, and taking the association relationship between the plurality of graph nodes as edges. In an embodiment, the graph construction module 620 can be used to perform the operation S220 described in the foregoing, and details are not repeated here.

[0193] The strategy generation module 630 is configured to train the target graph to obtain an adjustment strategy of the parameter value of the target memory configuration parameter. In an embodiment, the strategy generation module 630 can be configured to perform the operation S230 described above, and details are not repeated here.

[0194] The parameter update module 640 is configured to update the parameter value of the memory configuration parameter stored in the shared memory according to the adjustment strategy, so that the basic input / output system reads from the shared memory and re-executes the memory training based on the updated parameter value. In an embodiment, the parameter update module 640 can be configured to perform the operation S240 described above, and details are not repeated here.

[0195] Figure 7 A structural block diagram of a memory training apparatus applied to a basic input / output system according to an embodiment of the present application is shown.

[0196] As shown in Figure 7 The memory training apparatus 700 applied to the basic input / output system of this embodiment includes a parameter reading module 710, a memory training module 720, a log transmission module 730, an update response module 740, and a re-training module 750.

[0197] The parameter reading module 710 is configured to read the parameter value of the memory configuration parameter from the shared memory, wherein the shared memory is accessed by the basic input / output system and the management controller. In an embodiment, the parameter reading module 710 can be configured to perform the operation S410 described above, and details are not repeated here.

[0198] The memory training module 720 is configured to perform the memory training based on the parameter value. In an embodiment, the memory training module 720 can be configured to perform the operation S420 described above, and details are not repeated here.

[0199] The log transmission module 730 is configured to transmit a training log representing the failure of the memory training to the management controller in the case of the failure of the memory training. In an embodiment, the log transmission module 730 can be configured to perform the operation S430 described above, and details are not repeated here.

[0200] The update response module 740 is configured to re-read the updated parameter value from the shared memory in response to the management controller completing the update of the parameter value in the shared memory. In an embodiment, the update response module 740 can be configured to perform the operation S440 described above, and details are not repeated here.

[0201] The re-training module 750 is configured to re-execute the memory training based on the updated parameter value. In an embodiment, the re-training module 750 can be configured to perform the operation S450 described above, and details are not repeated here.

[0202] According to embodiments of the present application, any of the log analysis module 610, the graph construction module 620, the policy generation module 630 and the parameter update module 640, and any of the parameter reading module 710, the in-memory training module 720, the log transmission module 730, the update response module 740 and the re-training module 750 can be combined in one module, or any of them can be split into multiple modules. Alternatively, at least part of the function of one or more of these modules can be combined with at least part of the function of other modules, and implemented in one module. According to embodiments of the present application, at least one of the log analysis module 610, the graph construction module 620, the policy generation module 630 and the parameter update module 640, and at least one of the parameter reading module 710, the in-memory training module 720, the log transmission module 730, the update response module 740 and the re-training module 750 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system on package, an application specific integrated circuit (ASIC), or any other reasonable way of integrating or packaging a circuit, etc. or implemented by hardware or firmware, or implemented in any one of software, hardware and firmware or in a proper combination of any of them. Alternatively, at least one of the log analysis module 610, the graph construction module 620, the policy generation module 630 and the parameter update module 640, and at least one of the parameter reading module 710, the in-memory training module 720, the log transmission module 730, the update response module 740 and the re-training module 750 can be at least partially implemented as a computer program module which, when executed, can perform the corresponding function.

[0203] Figure 8 A block diagram of an electronic device suitable for implementing the in-memory training method according to embodiments of the present application is shown.

[0204] As shown in Figure 8 The electronic device 800 according to embodiments of the present application includes a processor 801 which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 802 or loaded from a storage portion 808 into a random access memory (RAM) 803. The processor 801 can include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset, and / or a special-purpose microprocessor (e.g., an application specific integrated circuit (ASIC)), etc. The processor 801 can also include an on-board memory for cache use. The processor 801 can include a single processing unit or multiple processing units for performing different actions of the method processes according to embodiments of the present application.

[0205] In the RAM 803, various programs and data required for the operation of the electronic device 800 are stored. The processor 801, the ROM 802, and the RAM 803 are connected to each other via the bus 804. The processor 801 performs various operations of the method flow according to the embodiments of the present application by executing the programs in the ROM 802 and / or the RAM 803. It should be noted that the programs can also be stored in one or more memories other than the ROM 802 and the RAM 803. The processor 801 can also perform various operations of the method flow according to the embodiments of the present application by executing the programs stored in the one or more memories.

[0206] According to the embodiments of the present application, the electronic device 800 can further include an input / output (I / O) interface 805, which is also connected to the bus 804. The electronic device 800 can further include one or more of the following components connected to the input / output (I / O) interface 805: an input part 806 including a keyboard, a mouse, etc.; an output part 807 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 808 including a hard disk, etc.; and a communication part 809 including a network interface card such as a LAN card, a modem, etc. The communication part 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as necessary. A removable recording medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 810 as necessary, so that a computer program read out therefrom is installed in the storage part 808 as necessary.

[0207] The present application also provides a computer readable storage medium, which can be included in the device / apparatus / system described in the above embodiments; or can exist separately without being assembled into the device / apparatus / system. The above computer readable storage medium carries one or more programs, when the one or more programs are executed, the method according to the embodiments of the present application is implemented.

[0208] According to an embodiment of the present application, the computer readable storage medium can be a non-transitory computer readable storage medium, for example, can include but not limited to: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present application, the computer readable storage medium can include the ROM 802 and / or the RAM 803 described above and / or one or more memory other than the ROM 802 and the RAM 803.

[0209] Embodiments of the present application also include a computer program product, which includes a computer program containing program codes for executing the method shown in the flow chart. When the computer program product is run in a computer system, the program codes are used to make the computer system implement the memory training method provided by the embodiments of the present application.

[0210] The above functions defined in the system / device of the embodiments of the present application are performed when the computer program is executed by the processor 801. According to an embodiment of the present application, the system, device, module, unit, etc. described above can be implemented by computer program modules.

[0211] In one embodiment, the computer program can rely on tangible storage media such as optical storage media, magnetic storage media, etc. In another embodiment, the computer program can also be transmitted, distributed, downloaded and installed in the form of signals on a network medium, and downloaded and installed through the communication part 809, and / or installed from the detachable medium 811. The program codes contained in the computer program can be transmitted by any suitable network medium, including but not limited to wireless, wired, etc., or any suitable combination of the foregoing.

[0212] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the detachable medium 811. When the computer program is executed by the processor 801, the above functions defined in the system of the embodiments of the present application are performed. According to an embodiment of the present application, the system, device, apparatus, module, unit, etc. described above can be implemented by computer program modules.

[0213] According to embodiments of the present application, program code for implementing the computer programs provided by embodiments of the present application can be written in any combination of one or more programming languages, and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The programming language can include, but is not limited to, Java, C++, python, "C" language, or similar programming languages. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the remote computing device, or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computing device, such as through the Internet using an Internet Service Provider.

[0214] The computer program instructions can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0215] Those skilled in the art will appreciate that features recited in the various embodiments of the present application can be combined and / or integrated in various combinations and / or permutations, even if such combinations and / or permutations are not expressly noted in the present application. In particular, features recited in the various embodiments of the present application can be combined and / or integrated in various combinations and / or permutations without departing from the spirit and scope of the present application. All such combinations and / or integrations are within the scope of the present application.

[0216] The embodiments of the present application have been described above. However, these embodiments are merely for the purpose of illustration and are not intended to limit the scope of the present application. Although the respective embodiments are described above separately, this does not mean that the measures in the respective embodiments cannot be used advantageously in combination. Various alternatives and modifications can be made to the embodiments of the present application by those skilled in the art without departing from the scope of the present application, and all such alternatives and modifications are intended to fall within the scope of the present application.

Claims

1. A memory training method, characterized by, The method is applied to a management controller and comprises the following steps: In response to receiving a training log indicating a memory training failure transmitted by a basic input / output system, the training log is parsed to determine a target training phase in which the failure occurs and a training failure field describing a training failure state, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information indicating a failure of a parameter value setting of the target memory configuration parameter; A target graph is generated by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking an association relationship between the plurality of graph nodes as edges; The target graph is trained to obtain an adjustment strategy for a parameter value of the target memory configuration parameter; According to the adjustment strategy, the parameter value of the memory configuration parameter stored in a shared memory is updated, so that the basic input / output system reads from the shared memory and re-executes memory training based on the updated parameter value; The method further comprises the following steps: In a case where the target training phase includes a plurality of target training phases, a root cause phase causing the failure is determined from the plurality of target training phases by analyzing a time sequence relationship between the plurality of target training phases; A root cause graph is generated by taking the root cause phase, a root cause memory configuration parameter corresponding to the root cause phase, and state information associated with the root cause memory configuration parameter as graph nodes and taking an association relationship between the plurality of graph nodes as edges; The root cause graph is trained to obtain an adjustment strategy for a parameter value of the root cause memory configuration parameter.

2. The method of claim 1, wherein, The target graph is generated by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking an association relationship between the plurality of graph nodes as edges, and comprises the following steps: The target training phase, the target memory configuration parameter, and the state information are defined as a first-level graph node, a second-level graph node, and a third-level graph node, respectively; The first-level graph node and the corresponding second-level graph node, and the second-level graph node and the corresponding third-level graph node are connected based on a configuration association relationship between the target training phase and the target memory configuration parameter and a state association relationship between the target memory configuration parameter and the state information, to obtain the target graph.

3. The method according to claim 1 or 2, characterized in that, The target graph is trained to obtain an adjustment strategy for a parameter value of the target memory configuration parameter, and comprises the following steps: Feature extraction is performed on graph information in the target graph to obtain a target feature vector representing features of the target graph; The target feature vector is input into a decision module to process the target feature vector by the decision module to obtain the adjustment strategy.

4. The method of claim 3, wherein, The method further comprises the following steps: In response to receiving an update training log indicating a memory training failure based on the updated parameter value transmitted by the basic input / output system, a feedback signal indicating that the adjustment strategy is invalid is generated based on the update training log; Based on the feedback signal, the target graph, and the adjustment strategy are determined as negative training samples; The parameters of the decision module are updated using the negative training samples.

5. The method of claim 1, wherein, The adjustment strategy includes a parameter adjustment type and a parameter adjustment amount; and updating the parameter value of the target memory configuration parameter stored in the shared memory according to the adjustment strategy includes: Obtaining the current parameter value of the target memory configuration parameter from the shared memory; Determining an updated parameter value after adjusting the current parameter value according to the parameter adjustment type and the parameter adjustment amount, wherein the parameter adjustment type is increasing or decreasing; Comparing the updated parameter value with a preset parameter threshold corresponding to the target memory configuration parameter; In the case where the updated parameter value is within the preset parameter threshold, updating the parameter value of the target memory configuration parameter stored in the shared memory to the updated parameter value.

6. The method of claim 5, wherein, The updating of the parameter value of the target memory configuration parameter stored in the shared memory according to the adjustment strategy further includes: In the case where the updated parameter value is outside the preset parameter threshold, updating the parameter value of the target memory configuration parameter stored in the shared memory to a boundary parameter value closest to the updated parameter value among the preset parameter threshold.

7. The method of claim 1, wherein, The method further includes: Recording the number of repetitions of memory training for the same target memory configuration parameter by using a counter; In the case where the number of repetitions reaches a preset number threshold and memory training still fails, suspending the adjustment of the parameter value of the target memory configuration parameter, encapsulating the corresponding historical target graph, historical adjustment strategy, and historical training log as an abnormal data packet, and uploading the abnormal data packet to an operation and maintenance management platform.

8. The method of claim 1, wherein, Before generating the target graph by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking the association relationship between the plurality of graph nodes as edges, the method further includes: Matching in a local adjustment knowledge base by taking the target training phase, the target memory configuration parameter, and the state information as query conditions to generate a matching result, wherein the local adjustment knowledge base stores successful adjustment records, and the successful adjustment records include a mapping relationship between a training phase and a memory configuration parameter corresponding to the training phase; In the case where the matching result represents that there is a corresponding successful adjustment record, writing the parameter value in the successful adjustment record into the shared memory.

9. The method of claim 8, wherein, The method further includes: In the case where it is determined that memory training based on the updated parameter value is successful, establishing a mapping relationship between the updated parameter value currently stored in the shared memory and the updated parameter value of the target training phase and the target memory configuration parameter corresponding to the adjustment strategy triggering this successful training, generating a corresponding successful adjustment record, and storing the successful adjustment record into the local adjustment knowledge base.

10. The method of claim 1, wherein, The analysis of the training log includes: Determining a log line including a preset error identifier from the training log; extract a target memory configuration parameter associated with the preset error identifier from the log line, and read a value field and a state description field corresponding to the target memory configuration parameter; perform semantic recognition on the state description field to determine state information representing a failure of setting a parameter value of the target memory configuration parameter; determine the training failure field based on the target memory configuration parameter and the state information.

11. The method of claim 1, wherein, The method further comprises: In a case where the updating of the parameter value stored in the shared memory is completed, the basic input / output system sends a parameter ready signal, so that the basic input / output system, in response to receiving the parameter ready signal, restarts, re-reads the updated parameter value from the shared memory, and re-executes the memory training based on the updated parameter value.

12. A memory training method, comprising: The method applied to a basic input / output system comprises: reading a parameter value of a memory configuration parameter from a shared memory, wherein the shared memory is accessed by the basic input / output system and a management controller; performing memory training based on the parameter value; in a case where the memory training fails, transmitting a training log representing the failure of the memory training to the management controller to trigger the management controller to perform the following operations: in response to receiving the training log representing the failure of the memory training transmitted by the basic input / output system, analyzing the training log to determine a target training phase in which the failure occurs and a training failure field describing a state of the training failure, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing a failure of setting a parameter value of the target memory configuration parameter; generating a target graph by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking an association relationship between the plurality of graph nodes as edges; training the target graph to obtain an adjustment strategy for the parameter value of the target memory configuration parameter; updating the parameter value of the memory configuration parameter stored in the shared memory according to the adjustment strategy, so that the basic input / output system reads the updated parameter value from the shared memory and re-executes the memory training based on the updated parameter value; wherein, in a case where the target training phase includes a plurality of target training phases, a root cause phase causing the failure is determined from the plurality of target training phases by analyzing a time sequence relationship between the plurality of target training phases; generating a root cause graph by taking the root cause phase, a root cause memory configuration parameter corresponding to the root cause phase, and state information associated with the root cause memory configuration parameter as graph nodes and taking an association relationship between the plurality of graph nodes as edges; training the root cause graph to obtain an adjustment strategy for the parameter value of the root cause memory configuration parameter; in response to the management controller completing the updating of the parameter value in the shared memory, re-reading the updated parameter value from the shared memory; re-executing the memory training based on the updated parameter value. The method applied to a basic input / output system comprises: reading a parameter value of a memory configuration parameter from a shared memory, wherein the shared memory is accessed by the basic input / output system and a management controller; performing memory training based on the parameter value; in a case where the memory training fails, transmitting a training log representing the failure of the memory training to the management controller to trigger the management controller to perform the following operations: in response to receiving the training log representing the failure of the memory training transmitted by the basic input / output system, analyzing the training log to determine a target training phase in which the failure occurs and a training failure field describing a state of the training failure, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing a failure of setting a parameter value of the target memory configuration parameter; generating a target graph by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking an association relationship between the plurality of graph nodes as edges; training the target graph to obtain an adjustment strategy for the parameter value of the target memory configuration parameter; updating the parameter value of the memory configuration parameter stored in the shared memory according to the adjustment strategy, so that the basic input / output system reads the updated parameter value from the shared memory and re-executes the memory training based on the updated parameter value; wherein, in a case where the target training phase includes a plurality of target training phases, a root cause phase causing the failure is determined from the plurality of target training phases by analyzing a time sequence relationship between the plurality of target training phases; generating a root cause graph by taking the root cause phase, a root cause memory configuration parameter corresponding to the root cause phase, and state information associated with the root cause memory configuration parameter as graph nodes and taking an association relationship between the plurality of graph nodes as edges; training the root cause graph to obtain an adjustment strategy for the parameter value of the root cause memory configuration parameter; in response to the management controller completing the updating of the parameter value in the shared memory, re-reading the updated parameter value from the shared memory; re-executing the memory training based on the updated parameter value.

13. The method of claim 12, wherein, Before performing the reading of the parameter value of the memory configuration parameter from the shared memory, the method further comprises: reading a memory training result flag stored in the non-volatile memory; in a case where the memory training result flag is a first identification, performing the reading of the parameter value of the memory configuration parameter from the shared memory; in a case where the memory training result flag is a second identification, reading a historical successful parameter value from the non-volatile memory and performing memory training based on the historical successful parameter value.

14. The method of claim 13, wherein, After performing the re-execution of the memory training, the method further comprises: in a case where the memory training is successful, writing a parameter value used when the current training is successful into the non-volatile memory and updating the memory training result flag to the second identification.

15. The method of claim 13, wherein, Further comprising: in response to receiving a parameter ready signal sent by the management controller, performing a system restart operation; in a case where the restart is completed, reading an updated parameter value from the shared memory to re-perform memory training.

16. A memory training apparatus, comprising: The apparatus comprises: a log analysis module configured to, in response to receiving a training log representing a failure of memory training transmitted by a basic input output system, analyze the training log to determine a target training phase in which the failure occurs and a training failure field describing a state of the training failure, wherein the memory training sequentially passes through a plurality of training phases, the plurality of training phases respectively adjust different types of memory configuration parameters, and the training failure field includes a target memory configuration parameter and state information representing a failure of setting a parameter value of the target memory configuration parameter; a graph construction module configured to generate a target graph by taking the target training phase, the target memory configuration parameter, and the state information as graph nodes and taking an association relationship between the plurality of graph nodes as edges; a strategy generation module configured to train the target graph to obtain an adjustment strategy for the parameter value of the target memory configuration parameter; a parameter update module configured to update the parameter value of the memory configuration parameter stored in the shared memory according to the adjustment strategy, so that the basic input output system reads the updated parameter value from the shared memory and re-performs memory training based on the updated parameter value; the apparatus is further configured to, in a case where the target training phase includes a plurality of target training phases, determine a root cause phase causing the failure from the plurality of target training phases by analyzing a time sequence relationship between the plurality of target training phases, generate a root cause graph by taking the root cause phase, a root cause memory configuration parameter corresponding to the root cause phase, and state information associated with the root cause memory configuration parameter as graph nodes and taking an association relationship between the plurality of graph nodes as edges, and train the root cause graph to obtain an adjustment strategy for the parameter value of the root cause memory configuration parameter. 17.An electronic device, comprising: one or more processors; a memory configured to store one or more computer programs, characterized in that the one or more processors execute the one or more computer programs to implement steps of the method according to any one of claims 1-15.

18. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions implement the steps of the method according to any one of claims 1-15 when executed by a processor.

19. A computer program product, characterised in that, A computer program comprising instructions which, when executed by a processor, implement the method according to any one of claims 1-15.

Images