Reliability method for single event upset effect based on SoC chip
By introducing a multivariate reliability framework of PL, APU and RPU on SoC chips, combining heartbeat monitoring and three-mode redundancy mechanisms, the problem of insufficient soft error protection in the space radiation environment of SoC chips is solved, and the system's high reliability is achieved and reduce maintenance costs.
Patent Information
- Application Number
- CN202111411909.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-25
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-11-25
AI Technical Summary
The existing SoC chips lack effective soft error protection technology in space radiation environment, resulting in unreliable systems, and the existing protection solutions are single, lacking a unified reliability software development framework, which makes the system maintenance cost high.
Provide a reliability method framework based on SoC chip, including three application units: PL, APU and RPU, adopting APU error detection mechanism, APU recovery mechanism, RPU error detection mechanism and RPU recovery mechanism, combining heartbeat monitoring and three-mode redundancy mechanism to detect and protect soft errors.
Through multi-object protection technology, the soft error protection capability under the SoC architecture is improved, the system's reliability under the single-particle flip effect is ensured, and the system maintenance cost is reduced.
Smart Images

Figure CN114416436B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of high reliability of embedded circuits, and specifically relates to a reliability method based on a SoC chip facing a single particle upset effect. The present invention is a method based on a reliability architecture of a SoC chip in a space radiation environment, and belongs to a method for reliability work of an electronic chip in a circuit. Background Art
[0002] SoC chips refer to the integration of FPGA and multiple processor cores on a single chip. SoC chips allow users to customize logic function circuits. The internal heterogeneous multi-core processors can provide higher processor performance and more efficient power utilization at a lower main frequency, and occupy a smaller physical space. Therefore, they are widely used in various fields.
[0003] The improvement of SoC chip production technology will inevitably make the single particle effect of SoC more significant in the space radiation environment. The existing single particle protection scheme is usually aimed at FPGA, that is, the FPGA is equipped with fault-tolerant functional circuits to reduce the impact of soft errors on the system. Common fault-tolerant functional circuits include triple-mode redundant circuits. SoC contains multiple processors (PS) and FPGAs (PL). Under the SoC architecture, the common FPGA fault-tolerant circuit is transformed into a soft error detection circuit arranged in the FPGA. The PS end is responsible for identifying soft errors and generating corresponding soft error protection strategies. The change of protection method releases the logic resources in the FPGA, and takes advantage of the processor to handle complex tasks, giving full play to the advantages of SoC device resources to achieve the expected performance. There are six Arm cores and FPGAs in the RFSoC device. The six Arm cores are composed of APU (Application Processing Unit) and RPU (Real-time Processing Unit) respectively. Although the advantages of the processor are obvious, the processor may still have soft errors, which leads to unreliable application user functions; in addition, the number of processor cores is increased, and there is a lack of a unified reliability software development framework for soft error protection, and the system maintenance cost is high. Summary of the invention
[0004] The purpose of the present invention is to solve the shortcomings of the existing soft error protection technology, which is single-oriented and lacks soft error protection technology in the SoC system, and thus provide a reliability method framework based on the single particle upset effect of the SoC chip.
[0005] To achieve the above object, the present invention is implemented through the following technical solutions:
[0006] The embodiment of the present invention provides a reliability method based on a SoC chip facing a single-particle upset effect. The software reliability framework includes three application units: PL, APU, and RPU.
[0007] The PL is used to perform the task of detecting soft errors occurring in the PL;
[0008] The APU includes four processor cores and uses an APU error detection mechanism and an APU recovery mechanism to perform soft error identification tasks occurring in the PL;
[0009] The RPU includes two processor cores, and adopts an RPU error detection mechanism and an RPU recovery mechanism to perform the task of generating PL protection reinforcement technology for identifying soft errors.
[0010] In the above scheme, the software reliability framework also includes: APU error detection mechanism, APU recovery mechanism, RPU error detection mechanism and RPU recovery mechanism;
[0011] The APU error detection mechanism is used to detect soft errors occurring in the APU, including soft errors of operation results and soft errors occurring during operation;
[0012] The APU recovery mechanism is used to recover system failures caused by soft errors in the APU;
[0013] The RPU error detection mechanism is used to detect soft errors occurring in the RPU, i.e., a lock-step mechanism;
[0014] The RPU recovery mechanism is used to recover system failures caused by soft errors in the RPU, namely, a secondary rollback recovery mechanism and system reset.
[0015] In the above scheme, the APU error detection mechanism also includes: heartbeat monitoring rules, heartbeat monitoring module, operation result sending module, and heartbeat judgement module;
[0016] The heartbeat monitoring rule is used to divide the heartbeat monitoring module;
[0017] The heartbeat monitoring module is used to monitor the health status of the triple-module redundant execution core and generate heartbeat information of the triple-module redundant execution core in the APU;
[0018] The operation result sending module is used to send the operation result and heartbeat information of the triple modular redundant execution core in the APU to the heartbeat judgement module in the RPU;
[0019] The heartbeat determiner module is used to compare the heartbeat information or operation results using a three-choice-two mechanism to obtain correct heartbeat information or operation results.
[0020] In the above scheme, the heartbeat monitoring rules include basic block division rules, heartbeat monitoring point setting rules, and heartbeat information selection rules;
[0021] The basic block partitioning rule is used to partition the single-particle effect minimum monitoring unit;
[0022] The heartbeat information selection rule is used to select variables and other information to form the heartbeat information;
[0023] The heartbeat monitoring point setting rule is used to set the location where the heartbeat information pops up.
[0024] In the above scheme, the heartbeat monitoring module includes a basic block, a heartbeat monitoring point and heartbeat information;
[0025] In the above scheme, the heartbeat determiner module includes a mode selection function, a data comparison function, a recorder function and a fault and scheme matching function;
[0026] The mode selection function is used to adjust the working mode of the heartbeat determiner and compare the operation result or the heartbeat signal;
[0027] The data comparison function is used to compare the received operation result and the heartbeat signal in a three-choice-two mechanism, select the correct result, and describe the fault model;
[0028] The recorder function is used to record the number of working cores, the number of times each core of the APU participates in work and the number of abnormal working times, and calculate the core failure frequency f Fault ;
[0029] The fault and solution matching function is used to match the fault recovery solution according to the fault model.
[0030] In the above scheme, the mode selection function also includes that in the triple-mode redundant working mode, the heartbeat judge compares the operation results and selects the correct result to point to the RPU; in the heartbeat monitoring working mode, the heartbeat judge compares the heartbeat signal and records the faulty core, and the output of the heartbeat judge is sent back to the APU by the RPU.
[0031] In the above scheme, the APU recovery mechanism includes three triple-module redundant execution cores and one backup core;
[0032] The triple-module redundancy execution core is used to execute the working core of the triple-module redundancy mechanism;
[0033] The backup core is used when a triple-module redundant execution core fails. The backup core takes over the work of the failed core. At the same time, the backup core is converted into a triple-module redundant execution core, and is converted into a backup core after the failed core recovers.
[0034] In the above scheme, the reliability method further includes: the RPU error detection mechanism triggers the RPU recovery mechanism signal;
[0035] The RPU error detection mechanism triggers the RPU recovery mechanism signal, which is used to send a signal to enable the RPU recovery mechanism after a soft error triggers the RPU error detection mechanism.
[0036] Compared with the prior art, the present invention has the following advantages: the protection technology is transformed from a single object to a multiple object, and a reliability error correction method is provided under the SoC architecture to better protect against soft errors caused by single particle upset effects. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 is a schematic diagram of a data pipeline for SoC software and hardware collaborative design according to an embodiment of the present invention;
[0038] Figure 2 This is a reliability development framework diagram for hardware-software collaborative design of SoC for single-particle upset effects according to an embodiment of the present invention;
[0039] Figure 3 is a flow chart of a reliability reinforcement method for SoC according to an embodiment of the present invention;
[0040] Figure 4 is a schematic diagram of APU protection according to an embodiment of the present invention;
[0041] Figure 5 is a schematic diagram of basic block division of an APU heartbeat monitoring point according to an embodiment of the present invention;
[0042] Figure 6 is a schematic diagram of an APU heartbeat determiner according to an embodiment of the present invention;
[0043] Figure 7 Schematic diagram of a lockstep mechanism and rollback recovery according to an embodiment of the present invention. DETAILED DESCRIPTION
[0044] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the accompanying drawings by means of a specific implementation of heartbeat monitoring. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0045] Existing SoC chips usually combine FPGA with multi-core processors, taking advantage of the FPGA's parallel processing capabilities and custom logic functions and the control capabilities of multi-core processors to reduce power consumption and improve computing power. For example, Xilinx's SoC chip contains FPGA modules, quad-core Cortex-A53 modules and dual-core Cortex-R5 modules. In recent years, with the improvement of chip production and manufacturing processes, the multi-particle flip effect of chips in space radiation environments has gradually emerged. At present, relevant technicians have proposed a variety of protection measures for FPGA single-particle flip effects, and SoC chips are showing a trend of heterogeneous multi-core processors. Due to the existence of the processor's own single-particle flip effect, the reliability of PS-side decisions under the SoC architecture has been questioned. At present, there is still a lack of reliability protection technology under the SoC architecture and nuclear self-test technology under the SoC architecture.
[0046] The invention proposes a reliability method based on a SoC chip facing a single-particle upset effect.
[0047] The present invention designs a reliability method framework based on FPGA, quad-core Cortex-A53 processor and dual-core Cortex-R5 processor. The present invention requires the PL end to have soft error detection capability, and soft error identification and protection reinforcement strategy generation are required on the PS end. The present invention proposes a dynamic triple-mode redundancy mechanism on the PS end, and uses the quad-core Cortex-A53 to form a triple-mode redundancy and backup core mechanism to ensure the reliability of the APU processing unit. The present invention proposes a heartbeat error detection technology, which combines the APU and RPU to ensure the reliability of the processing core in the APU.
[0048] The PS-side kernel recovery technology of the present invention uses hot start for the faulty core of the APU, and the lock-step mechanism recovery of the RPU uses a two-level rollback recovery technology. The APU cores of the present invention communicate through the inter-core communication service, the RPU and the APU communicate using the inter-processor communication service, and the PL and the PS communicate using the AXI bus. Among them, the specific implementation methods of shared memory, inter-core interrupts, inter-processor interrupts and hot start are known to those skilled in the art and are not limited in this specification.
[0049] Figure 1 This is a data pipeline diagram for SoC hardware and software co-design. Figure 1 As shown in the figure, in a task cycle, the PL passes data to the APU for error identification, and the APU passes the error identification result to the RPU. In order to reduce the impact of data flow in the PS on the PL logical function, the PL will continue to execute the next cycle task while the PS is processing the PL data, reducing the system interrupt waiting time.
[0050] Figure 2The present invention is based on the hardware and software reliability method framework for single-particle upset effects of SoC chips, and is used for collaborative design of reliability development framework diagram. Figure 2 As shown in the figure, four A53 cores form a dynamic three-module redundancy mechanism (Dynamic TMR), and two R5 cores form a lock-step mechanism (Lock-Step). The lock-step mechanism requires the underlying hardware support of the processor core, while the dynamic three-module redundancy mechanism uses software to process data streams to constrain the core set. The dynamic three-module redundancy mechanism composed of four cores adopts the "3+1" mode. Three cores are responsible for forming a normal three-module redundancy mechanism, and one core is used as a hot backup to replace the faulty core in the three-module redundancy mechanism in time, so this core is also called the "backup core". The three-module redundancy mechanism ensures that the data results passing through this part of the processor core are normal. The lock-step mechanism is composed of two cores. The lock-step mechanism requires hardware support. When the reliability method program user function runs to a node, it performs data comparison. If and only if the comparison results of the two cores are the same, the program user function runs to the next step.
[0051] Continue to refer Figure 2 , the data flow of the software reliability framework is as follows: after the PL detects a soft error, it is passed to the L2 cache of the APU through the AXI_ACP interface in the SoC chip, and the APU is notified to process the data through an interrupt. The L2 cache has an ECC function and has a certain soft error protection capability. The calculation results of the triple-module redundancy in the APU are sent to the RPU in the form of inter-processor interrupts in turn, and the message buffer between processing units is used as a temporary storage location for the calculation results. After receiving the results, the RPU first calculates whether the results of the triple-module redundancy are consistent. If the three-choice two condition is not met, the RPU notifies the APU to recalculate the data of the cycle; if the three-choice two condition is met, the RPU calculates the protection strategy under the lockstep mechanism and reconfigures the PL through the PACP port.
[0052] Continue to refer Figure 2The PS error detection mechanism under the framework of software reliability method is the heartbeat monitoring technology of APU and the lock-step mechanism of RPU. The heartbeat monitoring technology in APU is to insert a heartbeat monitoring point in the user function. When the method program user function runs to the heartbeat monitoring point, it actively outputs a signal to the outside. The content of the heartbeat signal includes the data information of the running position and the current position, which is sent to RPU through IPI. The heartbeat judger located in the RPU lock-step mechanism is used to judge whether the triple-module redundant heartbeat signal meets the conditions. If the conditions are met, APU continues to execute the task. If it is judged that there is a problem with the a53 core at this time, the corresponding protection measures are executed. The lock-step mechanism of RPU is completed with hardware support. By comparing and comparing the calculation results at each step, the correctness of the calculation results is guaranteed. Therefore, it is not only a supporting technology for the reliability of data calculation results, but also a processor core error detection technology. Unlike the traditional lock-step mechanism, when the lock-step mechanism has a problem, the rollback recovery signal is triggered first instead of the reset signal. RPU also supports watchdog to ensure the reliability of RPU.
[0053] Continue to refer Figure 2 The PS recovery mechanism under the software reliability framework is the dynamic three-module redundancy technology of APU and the two-level rollback recovery technology of RPU. The dynamic three-module redundancy recovery technology refers to the technology that when the heartbeat judge determines that the core is in a faulty state, the backup core takes over the faulty core to ensure that the three-module redundancy is not interrupted. The two-level rollback recovery technology refers to the technology that saves the correct data of the two previous lock-step comparisons of the node in the rollback data backup area. When the comparison result of the node is wrong, the RPU is allowed to roll back to the data state of the previous two times. When the heartbeat judge located in the RPU determines that there is a problem with the three-module redundancy heartbeat signal and needs to switch the faulty core, the faulty core loses the ability to actively send the heartbeat signal and executes protection measures such as waiting for the next cycle task and hot start. Since the backup core is a hot backup mechanism, the heartbeat signal of the backup core is enabled at this time to determine whether the heartbeat of the backup core is consistent with the other two cores. If it is consistent, a dynamic three-module redundancy mechanism is formed. If it is inconsistent, the task of the cycle is recalculated. The lock-step mechanism of the RPU will perform redundant backup at each checkpoint and store it in the lock-step redundant backup area in the OCM. When the RPU needs to be rolled back for recovery, it obtains the backup data from the lock-step redundant backup area. When both rollbacks fail, the RPU reset signal is triggered. QSPI stores the startup files of the R5 core and the A53 core. OCM divides the lock-step redundant backup area and the software function areas corresponding to the RPU and APU. In addition to the above recovery technologies, both the APU and RPU support the bottom-up measures for function reset.
[0054] Figure 3 This is a flow chart of a reliability reinforcement method in a software and hardware reliability method framework based on a SoC chip facing a single-particle upset effect according to an embodiment of the present invention. Figure 3As shown, after the PL detects an error in the present invention, the data is sent to the APU for the APU to identify the error. In the process of identifying the error, the APU will continuously perform heartbeat monitoring to confirm whether a soft error occurs in the APU. If a soft error occurs in the APU, the APU itself has a corresponding recovery technology to ensure the normal operation of the APU itself. Under the premise that there is no problem with the heartbeat monitoring, the APU generates a correct PL error identification result and sends it to the protection reinforcement strategy unit of the RPU. In the process of the RPU calculating the reinforcement strategy, the RPU lock-step mechanism will simultaneously perform the RPU's own soft error detection to ensure that the data results of each step are consistent. If the lock-step comparison result is wrong, the corresponding measures will be executed to restore the RPU state. The heartbeat monitoring and lock-step mechanism ensure that the PS will eventually generate the correct PL protection strategy to ensure the reliability of the entire system.
[0055] Figure 4 Schematic diagram of APU protection according to an embodiment of the present invention. Figure 4 As shown in the figure, the data is sent to the L2 cache of the APU through S_AXI_ACP_FPD. In the APU, the four cores share the L2 cache, and the four cores read the data stored in the L2 cache in sequence. Although all four cores have obtained the data, when taking the calculation results, only three calculation results are taken, and one of the cores is used as a hot standby core. Figure 4 In the APU, CPU0, CPU1, and CPU2 form a triple-module redundancy mechanism, and CPU3 serves as a hot standby core.
[0056] The data processing results of the dynamic triple-module redundancy mechanism are sent to the RPU using the inter-processor interrupt mechanism. Each APU core initiates an inter-processor interrupt (IPI) to the RPU core in turn. The APU core writes a 32B request message in the message buffer, writes 1 to the bit of the target processor in the interrupt trigger register, and verifies whether the interrupt is issued by reading the observer register. Poll the observer register until the status bit is cleared, indicating that the receiver has processed the interrupt. The three cores in the triple-module redundancy mechanism send IPI interrupt requests to the target processor in turn.
[0057] The RPU is in lockstep at this time. When the processor in the RPU that is mainly responsible for the output receives the IPI interrupt from the APU, the RPU polls the status register to accept the interrupt processing request from the APU. When an interrupt is received, the 32B data from the APU is written into a variable x in the three-to-two voter, and a signal is sent to the sender by clearing the status register, indicating that the interrupt has been processed.
[0058] The lock-step state of the RPU comes from the support of the hardware, where the R5_0 processor is the main processor under the lock-step mechanism, and the R5_1 processor is the comparison core of the lock-step mechanism. The lock-step mechanism is an operation familiar to technicians in this field. The data results generated by the RPU can be sent to the PL via the AXI bus, and the PL can also be reconfigured through PCAP, including partial reconfiguration and dynamic reconfiguration.
[0059] Continue to refer Figure 4 , the heartbeat monitoring uses the same data channel as the operation data to be transmitted to the voter in the RPU. The triple-module redundant heartbeat monitoring algorithm inserts a heartbeat monitoring point into the user function. When the user function runs to the heartbeat monitoring point, the processor core actively sends a heartbeat signal to the lock-step core through an inter-core interrupt. The heartbeat signal includes the running position and the data variables of the current position. Since the heartbeat signal pops up during the running phase, it can effectively avoid and detect kernel crashes and runaways in advance.
[0060] Before the heartbeat is triggered, the value of the Boolean variable working is queried. Working = 1 indicates that the kernel is in the triple-module redundancy mechanism and a heartbeat should be triggered to be sent to the RPU; working = 0 indicates that the kernel is in the backup state and no heartbeat is triggered to be sent to the RPU. After the Boolean variable working is initially set, the RPU dynamically adjusts the kernel's working state based on the heartbeat information.
[0061] The judgment of the heartbeat signal is to use the three-choice-two mechanism to vote for the correct operation result after receiving the signals from the three cores. The heartbeat signal fault model is divided into the faulty core position correct data error, the faulty core position error, and the heartbeat signal of the faulty core is not received within the specified time. When the faulty core position is correct but the data is wrong, it indicates that the faulty core has a data flow error; when the faulty core position is wrong and the faulty core does not pop up the heartbeat signal, it indicates that the faulty core has a control flow error.
[0062] The heartbeat basic block is established according to the following rules.
[0063] Rule 1: When jump statements, branch statements, and call statements are not inside loop statements, user code is divided into method control flow basic blocks based on jump statements, branch statements, and call statements.
[0064] Rule 2: When global variables and user-defined key variables are not inside a loop statement, data flow basic blocks are refined based on the control flow basic blocks and with global variables and user-defined key variables as boundaries.
[0065] Rule 3: For loop statements, basic blocks should be divided after the loop statement ends, and the rules for dividing basic blocks by jump, branch, call statement and key variable should be ignored, that is, Rules 1 and 2 are invalid in loop statements. For loop statements with embedded loops, basic blocks are divided based on the "}" of the outermost loop statement.
[0066] Rule 4: According to whether the loop statement basic block contains jump statements, it can be divided into basic blocks with branch functions and basic blocks without branch functions. Among them, the basic blocks without branch functions only need to divide the loop statements into basic blocks, and the basic blocks with branch functions are further divided into b based on the loop statement basic blocks. ij , where i represents the basic block number of the loop statement and j represents the different branches of the basic block.
[0067] Rule 5: Except for the basic block containing loop statements, the remaining basic blocks cannot contain jump statements, branch statements and call statements except the last instruction.
[0068] Rule 6: Except for loop statement basic blocks, other basic blocks except the first instruction cannot be used as target instructions.
[0069] Figure 5 Schematic diagram of a heartbeat monitoring point APU according to an embodiment of the present invention. Figure 5 As shown in the figure, the code is divided into three basic blocks b0, b1, and b2. According to the changes of the intermediate variables, the b1 basic block will display different labels and be marked as different basic blocks b. 11 , b 12 , b 13 .
[0070] Heartbeat monitoring point setting rules: The label is set to a unique value corresponding to the current basic block, and its position should be placed at the end of the basic block and before the jump statement. For the loop statement branch basic block, its label is determined according to its final orientation. The heartbeat monitoring point should be set after the label.
[0071] When a basic block contains a loop statement with jump statements and branch statements, the loop statement has a branch function, and a variable that can represent the branch result is selected as the variable information, or a flag++ statement is added before the "}" of the outermost loop statement to indicate the number of times the loop is fully executed, with the flag signal as the key signal. The flag signal should be included in the calculation when calculating the basic block of the loop statement.
[0072] Heartbeat information selection rules: The heartbeat consists of the basic block label and the key variables in the basic block. Its basic composition is the core number Core, the basic block label Signature and the key variable keyWords. The data basic block selects the key variables to form the heartbeat signal. When the control basic block does not contain the key variables, the flag variable is selected or set as the variable information and the label to form the heartbeat signal.
[0073] After the heartbeat judge receives the heartbeat signals sent by the three processing cores, it compares the basic block labels Signature and key variables keyWords of the three cores in the heartbeat information of the three inner cores according to the three-choose-two mode, selects two or more identical data as the result output, and records the error core according to the core number Core.
[0074] Figure 6 Schematic diagram of an APU heartbeat determiner according to an embodiment of the present invention. Figure 6 As shown, the heartbeat detector has three input channels, namely, HB A , HB B , HB C , responsible for receiving the heartbeat information and final calculation results from the triple-module redundant working core.
[0075] The heartbeat judge has a heartbeat mode and a calculation result voting mode. The heartbeat mode means that the three-choice-two voter votes on the basic block label Signature and the key variable keyWords, records the error core, and feeds back the error to the APU. The calculation result voting mode means that the three-choice-two voter votes on the key variable keyWords, records the error core, calculates the failure rate of each core, and feeds back the error to the APU. The three-choice-two voter implements the data comparison function of the heartbeat judge.
[0076] The input channel mode_judgement is the heartbeat judgement mode judgment signal, which realizes the mode selection function of the heartbeat judgement.
[0077] The heartbeat judge has four outputs, namely the correct result true_result, fault mode fault_mode, fault core fault_core, and fault frequency fault_frequency. The correct result true_result is the correct result voted by the voter; the fault mode fault_mode analyzes the cause of the fault core; the fault core fault_core is the core with the fault, and the fault frequency fault_frequency is the result that needs to be output in the operation result voting mode, which refers to the fault occurrence rate corresponding to the four cores. The output of the heartbeat judge for each task cycle is recorded in the OCM on-chip storage area in the PS to realize the recorder function of the heartbeat judge.
[0078] Correspondence table between heartbeat and fault model
[0079]
[0080]
[0081] The above table is a correspondence diagram between the heartbeat and the fault model according to an embodiment of the present invention. The kernel faults that occur correspond to the coping situations one by one, that is, the heartbeat judgement fault and solution matching function.
[0082] At the same time, the failure rate f Fault It will also be used as a criterion for evaluating failures, with n being the number of kernel data or result errors and N being the number of times the kernel participates in the triple-module redundancy mechanism.
[0083]
[0084] When the failure rate is greater than 50%, it indicates that a single event upset has occurred in the initial data storage area of the core and a hot start needs to be performed.
[0085] Fault recovery is a key technology to ensure the normal operation of the dynamic three-module redundancy mechanism. The three-module redundant working core of the APU continuously sends heartbeat signals to the heartbeat judge located in the RPU. Continuing to refer to the above table, when the heartbeat judgement fault_mode variable indicates that fault one has occurred, and the fault_core variable indicates the faulty core, switch the system backup core, and switch the heartbeat signal switch of the faulty core and the backup core. When the fault_mode variable indicates that fault three has occurred, and the fault_core variable indicates the faulty core, switch the system backup core, switch the heartbeat signal switch of the faulty core and the backup core, and perform an A53 error core hot start. When fault_mode shows that faults two and four have occurred, recalculate the tasks of this cycle. When fault_mode shows fault five, the APU needs to be restarted.
[0086] The APU recalculates the task of this cycle when the RPU issues an instruction and the APU jumps back from the current instruction to the state when the PC register is 0.
[0087] After the hot-start core is started, it enters a waiting state. When the RPU receives the calculation result of the cycle and gives a feedback signal, it begins to be incorporated into the dynamic triple-module redundancy mechanism.
[0088] Figure 7 Schematic diagram of lockstep mechanism and rollback recovery according to an embodiment of the present invention. Figure 7 As shown in the figure, R5_0 and R5_1 execute the lockstep mechanism, and the input signal is transmitted in the form of IPI. When there is no error in the lockstep checkpoint, the data of this stage is backed up and stored in the OCM checkpoint. When an error occurs in the lockstep checkpoint, the rollback operation can restore CPU0 and CPU1 to the previous lockstep checkpoint, ensuring the high reliability of the processor core itself as a higher level. The watchdog timer is a supplementary error detection measure of the RPU under the lockstep mechanism. When the rollback fails and the watchdog reports an error, the system soft restart is triggered. The correct protection reinforcement strategy generated by the RPU is sent to PCAP via the AXI bus.
[0089] The rollback recovery design is a synchronous fault recovery design implemented on R5_0 and R5_1 after the fault is detected in lockstep. The backup data is stored in a first-in-first-out circular queue with a queue of 2, which means that two correct backup orders are stored. When rolling back and restoring, read from the end of the queue without deleting data. If the two backup data are still invalid, the system reset is triggered.
[0090] Each lock-step checkpoint is compared. If the results of the lock-step comparison are the same, the data information and status information of the checkpoint are stored in the rollback storage area. The rollback storage area stores two copies of the lock-step checkpoint information to achieve secondary rollback. When the secondary rollback fails, the RPU triggers the restart mechanism. During the RPU period, if the APU initiates an IPI interrupt communication to the RPU at this time, the APU needs to wait for the RPU to restart successfully. After the RPU restarts successfully, it sends a signal to the APU to notify the APU to send the interrupt communication again.
[0091] The watchdog timer is used to fill the gap in the RPU's lack of detection measures for situations such as user code running away and deadlock.
[0092] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.
Claims
1. A reliability method based on SoC chip facing single event upset effect, characterized by: The reliability method includes three application units: PL, APU (application processing unit), and RPU (real-time processing unit); The PL, i.e., the FPGA, is used to perform the task of detecting soft errors occurring in the PL; The APU includes four processor cores and uses an APU error detection mechanism and an APU recovery mechanism to perform soft error identification tasks occurring in the PL; The RPU includes two processor cores, and uses an RPU error detection mechanism and an RPU recovery mechanism to perform the task of generating PL protection reinforcement technology for identifying soft errors; The PL side has the ability to detect soft errors, and soft error identification and protection reinforcement strategy generation are required on the PS side; for the core recovery technology on the PS side, the faulty core of the APU adopts hot start, and the lock-step mechanism recovery of the RPU adopts the secondary rollback recovery technology; within a task cycle, the PL passes the data to the APU for error identification, and the APU passes the error identification result to the RPU; the calculation result of the triple-mode redundancy in the APU is sent to the RPU in the form of an inter-processor interrupt, and the message buffer between processing units is used as a temporary storage location for the calculation result; after receiving the result, the RPU first calculates whether the result of the triple-mode redundancy is consistent. If the condition of three-choice-two is not met, the RPU notifies the APU to recalculate the data of the cycle; if the condition of three-choice-two is met, the RPU calculates the protection strategy under the lock-step mechanism and reconfigures the PL through the PACP port; the PS error detection mechanism is the heartbeat monitoring technology of the APU and the lock-step mechanism of the RPU; the heartbeat monitoring technology in the APU The heartbeat monitoring technology inserts a heartbeat monitoring point into the user function. When the method program user function runs to the heartbeat monitoring point, it actively outputs a signal. The content of the heartbeat signal includes the data information of the running position and the current position, which is sent to the RPU through the IPI method. The heartbeat judge located in the RPU lock-step mechanism is used to judge whether the three-module redundant heartbeat signal meets the conditions. If the conditions are met, the APU continues to execute the task. If it is judged that there is a problem with the a53 core at this time, the corresponding protection measures are executed; when the lock-step mechanism has a problem, the rollback recovery signal is triggered first instead of the reset signal; the PS recovery mechanism is the dynamic three-module redundancy technology of the APU and the secondary rollback recovery technology of the RPU; the lock-step mechanism of the RPU will perform redundant backup at each checkpoint and store it in the lock-step redundant backup area in the OCM; when the RPU needs to roll back and recover, it obtains the backup data from the lock-step redundant backup area. When both rollbacks fail, the RPU reset signal is triggered.
2. The reliability method based on SoC chip facing single event upset effect according to claim 1, characterized in that: The reliability method includes: an APU error detection mechanism, an APU recovery mechanism, an RPU error detection mechanism and an RPU recovery mechanism; The APU error detection mechanism is used to detect soft errors occurring in the APU, including soft errors of operation results and soft errors occurring during operation; The APU recovery mechanism is used to recover system failures caused by soft errors in the APU; The RPU error detection mechanism is used to detect soft errors occurring in the RPU, i.e., a lock-step mechanism; The RPU recovery mechanism is used to recover system failures caused by soft errors in the RPU, namely, a secondary rollback recovery mechanism and system reset.
3. The reliability method based on SoC chip facing single event upset effect according to claim 1 or 2, characterized in that: The APU error detection mechanism also includes: heartbeat monitoring rules, heartbeat monitoring module, operation result sending module, heartbeat judgement module; The heartbeat monitoring rule is used to divide the heartbeat monitoring module; The heartbeat monitoring module is used to monitor the health status of the triple-module redundant execution core and generate heartbeat information of the triple-module redundant execution core in the APU; The operation result sending module is used to send the operation result and heartbeat information of the triple modular redundant execution core in the APU to the heartbeat judgement module in the RPU; The heartbeat determiner module is used to compare the heartbeat information or operation results using a three-choice-two mechanism to obtain the correct heartbeat information or operation results.
4. The reliability method based on SoC chip facing single event upset effect according to claim 3 is characterized in that: The heartbeat monitoring rules include basic block division rules, heartbeat monitoring point setting rules, and heartbeat information selection rules; The basic block partitioning rule is used to partition the single-particle effect minimum monitoring unit; The heartbeat information selection rule is used to select variable information to form heartbeat information; The heartbeat monitoring point setting rule is used to set the location where the heartbeat information pops up.
5. The reliability method based on SoC chip facing single event upset effect according to claim 3, characterized in that: The heartbeat monitoring module includes a basic block, a heartbeat monitoring point and heartbeat information.
6. The reliability method based on SoC chip facing single event upset effect according to claim 3, characterized in that: The heartbeat judgement module includes a mode selection function, a data comparison function, a recorder function and a fault and solution matching function; The mode selection function is used to adjust the working mode of the heartbeat determiner and compare the operation result or the heartbeat signal; The data comparison function is used to compare the received operation result and the heartbeat signal in a three-choice-two mechanism, select the correct result, and describe the fault model; The recorder function is used to record the number of working cores, the number of times each core of the APU participates in work and the number of abnormal working times, and calculate the core failure frequency f Fault ; The fault and solution matching function is used to match the fault recovery solution according to the fault model.
7. The reliability method based on SoC chip facing single event upset effect according to claim 6, characterized in that: The mode selection function also includes that in the triple-module redundant working mode, the heartbeat judge compares the operation results and selects the correct result to point to the RPU; in the heartbeat monitoring working mode, the heartbeat judge compares the heartbeat signal and records the faulty core, and the output of the heartbeat judge is sent back to the APU by the RPU.
8. The reliability method based on SoC chip facing single event upset effect according to claim 2, characterized in that: The APU recovery mechanism includes three triple-module redundant execution cores and one backup core; The triple-module redundancy execution core is used to execute the working core of the triple-module redundancy mechanism; The backup core is used when a triple-module redundant execution core fails. The backup core takes over the work of the failed core. At the same time, the backup core is converted into a triple-module redundant execution core, and is converted into a backup core after the failed core recovers.
9. The reliability method based on SoC chip facing single event upset effect according to claim 1, characterized in that: The reliability method framework also includes: RPU error detection mechanism triggering RPU recovery mechanism signal; The RPU error detection mechanism triggers the RPU recovery mechanism signal, which is used to send a signal to enable the RPU recovery mechanism after a soft error triggers the RPU error detection mechanism.
