CXL switch embedded fabric manager microkernel and communication protocol with host side management software

Abstract

The application discloses a CXL switch embedded fabric manager microkernel and a communication protocol with host-side management software, comprising: obtaining state information of a CXL switch port, performing a management operation on the port, and performing management communication with a host-side management entity. The embedded execution unit deployed in the CXL switch processes the port state information, and autonomously performs the management operation on the port according to the processing result. In the management communication, emergency management events and regular management events are distinguished, and the emergency management events are preferentially transmitted through a physical communication path independent of a CXL data plane. According to the availability of the host-side management entity, the management function range of the embedded execution unit is dynamically switched, and the cooperation of internal and external management entities is realized. The CXL switch has autonomous management capability, eliminates the single-point dependence on external management entities, ensures that emergency messages are delivered in time when the data plane is congested or fails, reduces the end-to-end response delay, and avoids configuration conflicts and management redundancy.

Description

Technical Field

[0001] This invention relates to a CXL Switch with an embedded Fabric Manager microkernel and a communication protocol with host-side management software, and particularly to the field of computer interconnection and system management technology. Background Technology

[0002] The CXL (Compute Express Link) Fabric Manager (FM) is the core management entity in the CXL architecture responsible for system composition and resource allocation. Existing CXL switch management typically relies on an external FM instance running on a host or BMC, with the FM managing and configuring devices through the Component Command Interface (CCI). However, this architecture has the following drawbacks:

[0003] As an external logical process, when the host carrying the FM crashes, the BMC fails, or the management network is interrupted, the entire CXLFabric will lose its management capabilities, and will be unable to respond to port hot-plug events, handle device errors, or perform dynamic capacity adjustments. The CXL switch itself lacks autonomous management capabilities.

[0004] The end-to-end path of management commands is too long. The typical path is: host-side management software → FM → MCTP / PCIe transmission → CXL switch CCI → command execution → response return. In a deeply cascaded topology, the latency is further increased, making it difficult to meet the sub-millisecond response requirements.

[0005] Management communications lack priority differentiation and QoS guarantees. When data plane traffic is congested, urgent management messages (such as error reports and hot-plug events) may be delayed in processing.

[0006] The boundaries of responsibility between embedded management entities and external management entities are blurred, and there is a lack of collaborative management mechanisms, which can easily lead to configuration conflicts or management redundancy. Summary of the Invention

[0007] This invention provides a CXL Switch embedded Fabric Manager microcore and a communication protocol with host-side management software to overcome the shortcomings of existing technologies, such as CXL switch management relying on external management entities, resulting in single point of failure, high response latency, lack of communication priority differentiation, and unclear responsibilities of internal and external management entities.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] This invention discloses a CXL Switch embedded Fabric Manager microkernel and a communication protocol with host-side management software, including: acquiring the status information of CXL switch ports; performing management operations on the ports; conducting management communication with the host-side management entity; processing the port status information using an embedded execution unit deployed inside the CXL switch, and performing management operations on the ports based on the processing results; in the management communication, distinguishing between emergency management events and regular management events, and prioritizing the transmission of emergency management events through a physical communication path independent of the CXL data plane; and dynamically switching the management function scope of the embedded execution unit according to the availability of the host-side management entity.

[0010] Furthermore, the embedded execution unit is an FM microcore deployed inside the CXL switch chip.

[0011] Furthermore, the step of dynamically switching the management function scope of the embedded execution unit based on the availability of the host-side management entity includes: in response to detecting that the host-side management entity is disconnected, expanding the management function of the FM microcore so that, in addition to its own basic management function, it can also respond to emergency management events and maintain the basic configuration of the CXL switch; in response to detecting that the host-side management entity recovers, shrinking the management function of the FM microcore so that it returns to the basic management function.

[0012] Furthermore, the detection of the host-side management entity being disconnected includes: based on periodic heartbeat messages, determining that the host-side management entity is disconnected when the number of consecutive heartbeat failures reaches a first threshold.

[0013] Furthermore, the extended management functions of the FM microcore also include: saving the current configuration state of the CXL switch as a configuration snapshot.

[0014] Furthermore, the physical communication path independent of the CXL data plane is a fast channel, which uses MCTP over I2C or MCTP over SMBus communication via a dedicated management port; the management communication also includes transmitting management events through a communication path that reuses the CXL data plane link, which is a standard channel, and the standard channel uses MMIO CCI or MCTP over PCIe communication via the data plane PCIe link.

[0015] Furthermore, the emergency management events include port failure notifications, AERR or FERR fatal error reporting, and hot-plug events; the routine management events include configuration queries, status readings, log collection, and performance statistics.

[0016] Furthermore, saving the current configuration state of the CXL switch as a configuration snapshot includes an event-driven triggering save operation, specifically including: monitoring the current state values ​​of multiple configuration entries; calculating the weighted sum of the state changes of each configuration entry between the current time and the previous time; triggering the saving of the configuration snapshot in response to the weighted sum of the state changes being greater than a dynamic threshold; and updating the dynamic threshold.

[0017] Furthermore, the sum of the weighted state changes The calculation formula is: ,in, To configure the total number of entries, For the first Preset weights for each configuration item, For the first The value of each configuration entry at the current moment. Its value at the previous moment; the dynamic threshold The update rule is: if Then save the snapshot and Reset to initial baseline threshold ;like Then, no snapshot will be saved, and... Updated to ,in This is a preset forgetting factor.

[0018] Furthermore, the first threshold is After the host-side management entity is determined to be disconnected, the response is that the number of consecutive successful heartbeats reaches greater than [a certain number]. The second threshold The host-side management entity is determined to be restored.

[0019] Furthermore, the step of prioritizing the transmission of the emergency management event through a physical communication path independent of the CXL data plane includes: calculating the health indices of the fast channel and the standard channel respectively; determining the transmission probability of selecting the fast channel and the standard channel for the management message to be sent based on its preset priority and the health index; and selecting the actual transmission communication path for the management message based on the transmission probability.

[0020] Furthermore, the health index The calculation formula is: ,in, Indicates channel identification. For the measurement delay of the corresponding channel, The preset maximum allowable delay, This represents the number of bit errors measured for the corresponding channel. The maximum allowed number of bit errors is preset; the transmission probability The calculation formula is: ,in, Indicates fast lane, Indicates standard channel, Indicates the priority of the management message. This is a sensitivity coefficient related to priority.

[0021] Furthermore, the management communication uses the MCTP protocol to encapsulate management messages.

[0022] Furthermore, the management operations include VCS configuration or port binding control of the CXL switch.

[0023] This application provides a CXL switch management device, which is deployed inside a CXL switch and includes: a status acquisition module for acquiring status information of CXL switch ports; an operation execution module for performing management operations on the ports; a management communication module for communicating with a host-side management entity; an embedded processing module for processing the port status information and driving the operation execution module to perform management operations based on the processing results; a communication scheduling module for distinguishing between emergency management events and regular management events during the communication process of the management communication module, and prioritizing the transmission of emergency management events through a physical communication path independent of the CXL data plane; and a role management module for dynamically switching the management function scope of the embedded processing module based on the availability of the host-side management entity.

[0024] This application provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the CXL Switch embedded Fabric Manager microkernel and the communication protocol with the host-side management software.

[0025] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the CXL Switch embedded Fabric Manager microkernel and the communication protocol with the host-side management software.

[0026] The beneficial effects achieved by this invention are as follows: By deploying an embedded execution unit inside the CXL switch, the switch is equipped with the ability to autonomously process port status information and perform management operations, thereby maintaining basic management even when the host-side management entity is disconnected, eliminating single-point dependence on external management entities; by distinguishing between emergency management events and regular management events, and prioritizing the transmission of emergency events through a physical communication path independent of the CXL data plane, it is ensured that emergency management messages can still be delivered in a timely manner when the data plane is congested or malfunctioning, significantly reducing end-to-end response latency; by dynamically switching the management function scope of the embedded execution unit according to the availability of the host-side management entity, the responsibilities of internal and external management entities are coordinated, avoiding configuration conflicts and management redundancy. Attached Figure Description

[0027] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0028] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0029] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0030] Example 1

[0031] Technical explanation:

[0032] FM microcore: refers to a lightweight, low-power hardware / firmware management core deployed inside the CXL switch chip. It can autonomously assume the basic management functions of the CXL Fabric when the external FM is unavailable, and has an independent power supply domain, clock source, and dedicated management memory. In this application, the FM microcore is a specific implementation of an embedded execution unit.

[0033] Fast lane: refers to a physical communication path independent of the CXL data plane (i.e., the PCIe link through which user data traffic passes), usually referring to MCTP over I2C or MCTP over SMBus communication links implemented through dedicated management ports on switches.

[0034] Standard Channel: Refers to the communication path that reuses the CXL data plane link. Management messages are carried on the data plane PCIe link, for example, through MMIO CCI (Component Command Interface) or MCTP over PCIe.

[0035] Configuration snapshot: refers to the periodic or event-driven backup and saving of the current complete configuration state of the CXL switch (such as port binding relationships, VCS configuration, routing table, etc.).

[0036] Role negotiation agreement: refers to the negotiation process between the FM microkernel and the host-side management entity to clarify their respective management roles (such as master role, guardian role) and responsibility boundaries, including interactive processes such as capability declaration, status reporting and responsibility division.

[0037] like Figure 1 As shown, a CXL Switch with an embedded Fabric Manager microkernel and a communication protocol with host-side management software includes:

[0038] S101: Obtain the status information of the CXL switch port;

[0039] S102: Perform management operations on this port;

[0040] S103: Perform management communication with the host-side management entity;

[0041] S104: The embedded execution unit deployed inside the CXL switch processes the port status information and performs management operations on the port based on the processing results;

[0042] S105: In this management communication, emergency management events are distinguished from regular management events, and the emergency management event is transmitted preferentially through a physical communication path independent of the CXL data plane;

[0043] S106: Dynamically switch the management function scope of the embedded execution unit based on the availability of the host-side management entity.

[0044] In this embodiment, the above steps work together to form a management closed loop with autonomy, high reliability, and low latency. Step S104 embeds an execution unit inside the switch, allowing management logic to be decentralized. Even if the host-side management entity (i.e., the external FM) is unavailable, the switch can still perform critical management operations based on local processing results, such as port fault isolation, eliminating the risk of single point of failure. Step S105 establishes an emergency communication path independent of the data plane, ensuring that notifications of emergency events such as port failures and hot-plugging can still be delivered to the host side with extremely low latency when the data plane link is congested or faulty, significantly improving the real-time performance of management response. Step S106 dynamically adjusts the management authority scope of the internal execution unit according to the online status of the external FM. It coordinates with the external FM when it is normal and autonomously expands its functions to maintain basic services when it is disconnected, achieving seamless integration of internal and external management and avoiding management chaos. The combination of these three elements completely changes the existing CXL management architecture's strong dependence on the external FM.

[0045] Taking a specific application scenario as an example, in an AI training server deployed with multiple levels of CXL switches, a port on a remote CXL switch suddenly experiences a physical layer failure. The embedded execution unit inside the switch (step S104) obtains the abnormal port status in real time. Since this event is classified as an emergency management event (port failure notification), the management entity prioritizes reporting the fault information to the host-side management software within sub-milliseconds via a fast channel independent of the data plane (step S105), such as through MCTP messages on the I2C bus. Simultaneously, during the brief interval when the host-side management software is temporarily unresponsive due to network jitter, the embedded execution unit, based on the processed abnormal status and following its current management function (or switching to extended functions if the external FM is unreachable), may autonomously perform port isolation operations (step S102) to prevent the fault from spreading. After the host-side management software restores communication, the embedded execution unit synchronizes the saved configuration snapshot and event logs to the external FM, enabling it to quickly regain awareness of the global state and negotiate subsequent resource reallocation strategies.

[0046] In one embodiment, the aforementioned embedded execution unit is an FM microcore deployed within the CXL switch chip. This FM microcore provides a reliable hardware foundation, and its power supply and clock design, independent of the data plane core, ensures that basic management functions continue to operate even in the event of a main chip logic failure. This approach elevates the reliability of management capabilities from reliance on external software processes to a level guaranteed by the chip's internal hardware.

[0047] In one embodiment, regarding the aforementioned dynamic switching of the management scope based on the availability of the host-side management entity (S106), this dynamic switching of the management scope of the embedded execution unit based on the availability of the host-side management entity specifically includes: in response to detecting that the host-side management entity is disconnected, expanding the management functions of the FM microcore so that, in addition to its own basic management functions, it can also respond to emergency management events and maintain the basic configuration of the CXL switch; in response to detecting that the host-side management entity recovers, shrinking the management functions of the FM microcore so that it returns to its basic management functions. Furthermore, based on the above principles, this approach can bring the beneficial effect of adaptive management flexibility. The system can automatically adjust the aggressiveness of the management strategy according to changes in the external management environment, maximizing availability while ensuring security.

[0048] In one embodiment, the detection of a host-side management entity becoming unreachable, as mentioned above, includes: determining that the host-side management entity is unreachable when the number of consecutive heartbeat failures reaches a first threshold based on periodic heartbeat messages. This approach provides a clear and quantifiable criterion for determining the unreachability of an external management entity, avoiding misjudgments based on a single timeout and ensuring the stability of state transitions.

[0049] In some embodiments, the extended management functions of the FM microcore, as mentioned above, further include saving the current configuration state of the CXL switch as a configuration snapshot. This approach provides a data foundation for rapid state synchronization during subsequent recovery by external management entities, significantly shortening the system recovery time (RTO). As a specific implementation, this configuration snapshot is stored in the Configuration Snapshot Area (CSA) within the dedicated management memory of the FM microcore.

[0050] In one embodiment, regarding the aforementioned physical communication path and management communication independent of the CXL data plane, this physical communication path is a fast channel, which employs MCTP over I2C or MCTP over SMBus communication via a dedicated management port. The management communication also includes transmitting management events via a communication path that reuses the CXL data plane link, which is a standard channel. This standard channel employs MMIO CCI or MCTP over PCIe communication via the data plane PCIe link. This approach enables the construction of a dual-channel management communication architecture with redundant paths. The fast channel ensures absolutely low latency for emergency events, while the standard channel provides high-bandwidth routine management capabilities when the data plane is normal, thus achieving a balance between latency and bandwidth and improving communication availability.

[0051] In some embodiments, the emergency management events and routine management events mentioned above include port failure notifications, AERR or FERR fatal error reporting, and hot-plug events; the routine management events include configuration queries, status readings, log collection, and performance statistics. This approach clearly defines event categories of different service importance, providing a clear classification basis for subsequent communication priority scheduling and resource allocation.

[0052] In one embodiment, the operation of saving the current configuration state of the CXL switch as a configuration snapshot, as mentioned above, includes an event-driven triggering save operation, specifically including: monitoring the current state values ​​of multiple configuration entries; calculating the weighted sum of the state changes of each configuration entry between the current time and the previous time; triggering the saving of the configuration snapshot in response to the weighted sum of state changes exceeding a dynamic threshold; and updating the dynamic threshold. This approach optimizes the configuration snapshot saving frequency from a fixed period to dynamic triggering based on the actual degree of configuration change, avoiding unnecessary storage operations when there are no changes or only minor changes. This significantly reduces write wear and computational overhead on dedicated management memory, extends hardware lifespan, and improves overall efficiency.

[0053] In some embodiments, for the aforementioned weighted sum of state changes and dynamic threshold update, the weighted sum of state changes... The calculation formula is: ,in, To configure the total number of entries, For the first Preset weights for each configuration item, For the first The value of each configuration entry at the current moment. Its value at the previous moment; this dynamic threshold. The update rule is: if Then save the snapshot and Reset to initial baseline threshold ;like Then, no snapshot will be saved, and... Updated to ,in This is a pre-defined forgetting factor. As a specific implementation method, The value range is [0,1], set by the system administrator according to the importance of the configuration item; the forgetting factor α ranges from [0.1, 0.3]; the initial baseline threshold... Usually set to the historical maximum. The value is 5%. By adopting a dynamic threshold combined with a forgetting factor update rule, it can maintain high trigger sensitivity during periods of frequent configuration changes, and gradually relax the trigger conditions during stable periods, thus achieving adaptive snapshot management.

[0054] In some embodiments, the first threshold mentioned above is... After the host-side management entity is determined to be out of contact, the number of consecutive successful heartbeats reaches greater than [a certain threshold]. The second threshold The system determines that the host-side management entity has recovered. This is achieved by setting a recovery determination threshold. Greater than the threshold for determining loss of contact This creates an asymmetric hysteresis window, which effectively prevents the management role from frequently switching between "guardian" and "coordinator" (i.e., state oscillation) due to network jitter or brief external FM restarts. This reduces unnecessary role negotiation message storms and improves the overall stability of the system.

[0055] In one embodiment, the aforementioned priority transmission of emergency management events via a physical communication path independent of the CXL data plane includes: calculating the health indices of the fast channel and the standard channel respectively; determining the transmission probability of selecting the fast channel or the standard channel for a management message to be sent based on its preset priority and the health indices; and selecting the actual transmission communication path for the management message based on the transmission probability. This approach upgrades the original simple "two-choice" hard switching to a "soft" probabilistic selection based on real-time channel quality, avoiding the "ping-pong effect" (frequent switching of messages between the two channels) that may occur when the health of the two channels is close to the critical value, ensuring stable message transmission and low latency, especially effective in guaranteeing the transmission of high-priority messages.

[0056] In some embodiments, the health index is calculated in relation to the aforementioned health index and the probability of transmission. The calculation formula is: ,in, Indicates channel identification. For the measurement delay of the corresponding channel, The preset maximum allowable delay, This represents the number of bit errors measured for the corresponding channel. The maximum allowed number of bit errors is preset; this transmission probability The calculation formula is: ,in, Indicates fast lane, Indicates standard channel, This indicates the priority of the management message. This is a sensitivity coefficient related to priority. As a specific implementation, for P0 / P1 level urgent messages, It can be set to 3 to amplify the advantages of the fast channel; for P2 / P3 level regular messages, It can be set to 1 for linear allocation. The health index calculation comprehensively considers latency and bit error rate, and can fully reflect the availability and reliability of the channel. Using this weighted probability selection method, when the fast channel is healthy, high-priority messages will almost certainly choose the fast channel; when the fast channel deteriorates, high-priority messages still have a probability of being sent through the still available standard channel, avoiding complete message blocking caused by single-point channel deterioration.

[0057] In some embodiments, the management communication uses the MCTP (Management Component Transport Protocol) to encapsulate management messages. As a specific implementation, the MCTP protocol ensures the standardization of management message formats, facilitating compatibility and interaction with host-side management software that conforms to the CXL specification.

[0058] In some embodiments, the management operation includes configuring the CXL switch (Virtual CXL Switch) or controlling port bonding. This approach clarifies the specific types of management tasks that the embedded execution unit can handle.

[0059] It should be noted that, in some optional implementations, the aforementioned "configuration entries" may specifically include port binding relationship entries, VCS configuration parameters, PBR (Port-Based Routing) routing table entries, or HDM (Host-managed Device Memory) decoder configurations, etc. The term "host-side management entity" typically refers to a Fabric Manager software instance running on the host operating system or BMC (Baseboard Management Controller). The term "heartbeat message" refers to a control message periodically sent by both the master and managed parties to confirm the other party's liveness status; its transmission also follows the aforementioned management communication protocol and channel. The "basic management functions" of the FM microcore may include port status monitoring, device enumeration, watchdog management, etc.; its extended functions may further include handling hot-plug events, performing dynamic capacity adjustments, and other more complex operations. The aforementioned dynamic responsibility switching process incorporates the role negotiation protocol; for example, after the external FM recovers, both parties will renegotiate to determine which specific guardian role the FM microcore should shrink to. In addition, in some alternative implementations, to improve the performance and robustness of each core feature, the following optimization details can be introduced: For the dedicated management memory of the FM microcore, a thermally sensitive partition layout can be implemented based on the internal temperature gradient of the chip, placing data for different functions (such as emergency codes and configuration snapshots) in SRAM partitions of different processes or voltages to optimize power consumption and reliability; For the dedicated management port of the fast channel, an overcurrent protection lock based on a MEMS bimetallic strip can be integrated, which physically disconnects the circuit in the event of a short circuit and can be controlled and reset by the FM microcore, improving hardware security; For the independent power supply domain of the FM microcore, waste heat generated by the data plane chip of the CXL switch can be used to provide milliwatt-level backup power by integrating a micro thermoelectric generator (TEG) module; For the management message buffer, a micro buffer structure driven by the microchannel thermal differential pressure can be designed between the FM microcore and the I2C controller to achieve extremely low-power priority message temporary storage using the chip's natural heat flow. In terms of state transition debouncing, in addition to setting an asymmetric hysteresis threshold, the heartbeat delay variance can be calculated based on a sliding window, and the heartbeat interval can be dynamically adjusted accordingly, reducing communication overhead while ensuring detection sensitivity. At the message scheduling level, a dynamic rate limiting and fallback mechanism based on queue depth and remaining message lifetime can be introduced for each priority queue to prevent low-priority messages from being "starved" by high-priority message storms. Regarding configuration synchronization, a timestamp can be bound to each configuration entry. By calculating the variance of the timestamp difference sequence, it can be determined whether incremental changes might be lost between two snapshots, thus intelligently selecting between full snapshots or incremental logs for synchronization, improving synchronization efficiency and consistency.

[0060] This application also provides a CXL switch management device, deployed inside a CXL switch, comprising: a status acquisition module for acquiring status information of CXL switch ports; an operation execution module for performing management operations on the ports; a management communication module for communicating with a host-side management entity; an embedded processing module for processing the port status information and driving the operation execution module to perform management operations based on the processing results; a communication scheduling module for distinguishing between emergency management events and regular management events during communication with the management communication module, and prioritizing the transmission of emergency management events through a physical communication path independent of the CXL data plane; and a role management module for dynamically switching the management function scope of the embedded processing module based on the availability of the host-side management entity. This device achieves management autonomy through the embedded processing module, ensures low latency and high availability of communication through the communication scheduling module, and realizes intelligent collaboration with external management entities through the role management module.

[0061] This application also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the CXL Switch embedded Fabric Manager microkernel and the communication protocol with the host-side management software.

[0062] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the CXL Switch embedded Fabric Manager microkernel and the communication protocol with the host-side management software.

[0063] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. The CXL Switch embeds a Fabric Manager microkernel and a communication protocol with the host-side management software, characterized in that: This includes obtaining the status information of the CXL switch ports; Perform management operations on the port; Perform management communication with the host-side management entity; The embedded execution unit deployed inside the CXL switch processes the port status information and performs management operations on the port based on the processing results. In the management communication, emergency management events are distinguished from regular management events, and emergency management events are preferentially transmitted through a physical communication path independent of the CXL data plane; The management function scope of the embedded execution unit is dynamically switched based on the availability of the host-side management entity.

2. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software as described in claim 1, characterized in that, The embedded execution unit is an FM microcore deployed inside the CXL switch chip.

3. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software as described in claim 2, characterized in that, The step of dynamically switching the management function scope of the embedded execution unit based on the availability of the host-side management entity includes: In response to the detection of the host-side management entity being disconnected, the management functions of the FM microcore are extended so that, in addition to its own basic management functions, it can also respond to emergency management events and maintain the basic configuration of the CXL switch; In response to the detection that the host-side management entity has recovered, the management functions of the FM microkernel are retracted, returning it to its basic management functions.

4. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software as described in claim 3, characterized in that, The detection of the host-side management entity being disconnected includes: based on periodic heartbeat messages, when the number of consecutive heartbeat failures reaches a first threshold, determining that the host-side management entity is disconnected.

5. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software according to claim 3 or 4, characterized in that, The extended management functions of the FM microcore also include: saving the current configuration state of the CXL switch as a configuration snapshot.

6. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software as described in claim 2, characterized in that, The physical communication path independent of the CXL data plane is a fast channel, which uses MCTP over I2C or MCTP over SMBus communication via a dedicated management port. The management communication also includes transmitting management events through a communication path of a multiplexed CXL data plane link. The communication path of the multiplexed CXL data plane link is a standard channel, which uses MMIO CCI or MCTP over PCIe communication via a data plane PCIe link.

7. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software as described in claim 1 or 6, characterized in that, The emergency management events include port failure notifications, AERR or FERR fatal error reports, and hot-plug events; The routine management events include configuration queries, status readings, log collection, and performance statistics.

8. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software as described in claim 5, characterized in that, Saving the current configuration state of the CXL switch as a configuration snapshot includes an event-driven save operation, specifically including: Monitor the current status values ​​of multiple configuration entries; Calculate the weighted sum of the state changes of each configuration item between the current time and the previous time. In response to the sum of the weighted state changes exceeding a dynamic threshold, a configuration snapshot is saved. Update the dynamic threshold.

9. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software according to claim 8, characterized in that, The sum of the weighted state changes The calculation formula is: , in, To configure the total number of entries, For the first Preset weights for each configuration item, For the first The value of each configuration entry at the current moment. Its value at the previous moment; The dynamic threshold The update rule is: if Then save the snapshot and Reset to initial baseline threshold ;like If so, a snapshot will not be saved, and will Updated to ,in This is a preset forgetting factor.

10. The CXL Switch embedded Fabric Manager microcore and communication protocol with host-side management software according to claim 4, characterized in that, The first threshold is After the host-side management entity is determined to be disconnected, the response is that the number of consecutive successful heartbeats reaches greater than [a certain number]. The second threshold The host-side management entity is determined to be restored.

Images