RU2437144C2 - Method to eliminate exception condition in one of nuclei of multinuclear system - Google Patents

Abstract

FIELD: information technologies.

SUBSTANCE: method includes stages, at which the following is carried out: cells are identified in the common memory, values of each nucleus are saved, and initial value of each nucleus condition is set as "normal"; when an exception situation occurs at one nucleus, it automatically performs the program of exception situation processing, sets the condition value as "exception situation" and notifies the other selected nucleus in the normal condition, then the nucleus in the exception condition automatically enters into the infinite cycle; the selected nucleus in normal condition sets the nucleus in exception condition into reset condition and notifies a system planning module, then this module transfers tasks that were initially assigned to the nucleus in exception condition, to the nucleus in normal condition for redistribution.

EFFECT: provision of system continuity and makes loss of system resources impossible before and after elimination of an exception situation in one nucleus.

5 cl, 1 dwg

Description

Technical field

The invention relates to a system with a multicore central processor (CPU) and, in particular, to a method for eliminating an exceptional situation in a single core of a multicore system.

BACKGROUND OF THE INVENTION

In systems with an integrated multi-core central processing unit (CPU) (hereinafter referred to as multi-core systems), exceptional situations can occur on a single core, regardless of whether the system is a symmetric multi-core system or a multi-core master-slave system. The term "exception" includes invalid commands, alignment errors, cache errors, data bus errors, etc. There are many reasons for exceptional situations. One of the reasons may be an accidental error of technical equipment or incorrect data that can cause an exceptional situation during the execution of the program, and another reason may be that the system executes a branch that is difficult to enter into the program. However, most of these errors simply cause one-time damage to the system, since systematic errors can be detected and corrected during system testing.

In well-known technical solutions, the usual approach to eliminating an exception that occurs on one core is simply to record all the information about the exception and then restart the entire system. Although this approach can restore the functioning of the system, it will pause the work of all services and reduce the working time of the system. For existing systems with a complex structure, in particular, they are generally placed at high-level positions or at the core position, such as a peripheral network core router, private telephone system, etc. As soon as equipment malfunctions, its consequences can be serious. In addition, restarting the system and restoring normal operation will take a long time, which can have enormous consequences. Therefore, it is especially important to extend the working time of a multi-core system. In addition, it makes no sense to restart the entire system just for the sake of some non-fatal errors.

The content of the invention

The technical problems that must be solved in the invention focus on the problems mentioned in the existing technology. The invention provides a recovery method for a single core in a multicore system. When an exception occurs in one core, that core will be restored without a pause in operation.

The technical scheme used in the invention to solve the mentioned technical problems consists in a method for eliminating an exceptional situation in one core of a multicore system, including shared memory and a system planning module.

The invention consists of the following steps.

but. Allocation of a shared memory cell, saving the state value of each core and setting the initial state value of all cores to “normal”.

b. When an exception occurs on one core, this kernel automatically executes the exception handling program, sets its state value to “exception” and notifies another selected kernel that is in a normal state, then the kernel in an exceptional state automatically enters an infinite loop.

from. The selected kernel in the normal state sets the kernel in the exceptional state to the reset state and notifies the system planning module. Then this module will transfer the tasks originally assigned to the kernel, which is currently in an exceptional state, to any other kernel, which is in a normal state. The selected kernel in the normal state will redistribute all resources occupied by the kernel, which is in the exceptional state, and, finally, free the kernel, which is in the exceptional state.

d. After the release from the reset, the kernel in an exceptional state will restart and set its state to the value “waiting for recovery” after the launch is completed.

e. After the selected kernel in the normal state detects that the value of the kernel in the exceptional state is “pending recovery”, it will set the state value of this exceptional kernel to “normal” and notify the system planning module.

In addition, in step b, an interrupt is used to send an alert to a multi-core information exchange system.

In addition, the system scheduling module will monitor the state of each core in accordance with the state value stored in the memory cell. If the module determines that the state of one core is exceptional, it will no longer assign tasks to this core, i.e. core in exceptional condition.

In particular, in the case where the multicore system is symmetric, in step b, the selected module in the normal state can be any module in this state.

In particular, in the case where the multicore system is a master-slave system, in step b, the selected core in the normal state is the core that is in the master state.

The invention has the following beneficial effects: when an exception occurs in one core of a multicore system, tasks initially assigned to a core that is currently in an exceptional state will be directed to another normal core to ensure that these tasks are completed on time and effectively ensure that the entire the system will not be suspended and system resources will not be lost before or after the kernel in exceptional condition is restored. After restoration, the kernel, which is in exceptional condition, can function normally, which increases the useful time of the system and increases its reliability.

Description of drawings

The drawing is a block diagram of a software embodiment of the invention.

Detailed Description of a Preferred Embodiment

The technical scheme of the invention is described in more detail in combination with the accompanying drawings and an embodiment.

For a multi-core system with shared memory and a system planning module, the invention defines a special cell in shared memory and allocates a global array for storing the state of each core. The index of this array can use the number of each core, and each element of the array corresponds to the state value of each core. All kernel state values are defined as “normal,” “exceptional,” and “pending recovery.” The initial state value of all nuclei is set to "normal". In a multi-core system, all tasks performed by each core are assigned by the system planning module. The state monitoring program is located in the system planning module. When the system planning module assigns tasks, first of all it will check the current state of each core and, if the state of one of the kernels is exceptional, the module will not assign tasks to this core. When an exception occurs on one of the cores, the CPU program that processes the exception will work.

In the program handling the exception according to the invention, the kernel in the exceptional state first sets its value to “exceptional”, and then selects the kernel in the normal state and uses the internuclear interrupt information exchange system to notify the selected kernel in the normal state. The system planning module then distributes all tasks assigned to the kernel in exceptional condition to another kernel in the normal state in accordance with its scheduling algorithm, so as to ensure, if possible, faster recovery and thereby reduce recovery time. After the notification is sent, the kernel in an exceptional state enters an endless loop and cannot exit the program that processes the exceptional state in order to prevent more errors and damage from occurring.

In a symmetric multicore system, any core can set the state of any other core. Therefore, when an exceptional state appears in one core, this core can select any other core in the normal state. In addition, any core has the function of resetting any other core or multiple cores. There are two algorithms for selecting one core in the normal state, namely sequential search and random search. The advantage of sequential search is simplicity, while its disadvantage is that the selected kernel in the normal state is relatively fixed. The advantage of random search is that the selected core in the normal state is not fixed, which can increase the likelihood of a successful recovery, while its disadvantage is that the algorithm is relatively complicated.

In a multicore master-slave system, only one core in a master state can restore a kernel in an exceptional state. In other words, when one core is in an exceptional state, it must notify the core in the “master” state so that it can perform a restore operation.

A multicore CPU (central processing unit) has internuclear information exchange mechanisms, of which one mechanism should use internuclear interruption. Its advantage lies in speed and in the fact that this mechanism can send an alert about an event in the first place. As a result, this invention uses an internuclear interrupt to send an alert.

Execution option

As shown in the drawing, in a symmetric system with an integrated multi-core CPU in step 101, an invalid operation appears in core A and causes an exception. Only kernel A goes to the branch of the exception and enters the CPU program that processes the exception, while other kernels continue to work normally. In a program that processes an exception, kernel A first writes information about the exception, including: type of exception, pointer to the program counter for the exception, the values of all state registers, the structure of the stack, etc.

In step 102, in the exception handling program, core A changes its state value in the shared memory cell to “exceptional”. When the system scheduling module assigns tasks, it first determines the state of each current kernel. If the current kernel is in exceptional condition, the module does not assign a task to this kernel.

In step 103, in an exception-processing program, kernel A randomly selects kernel B in the normal state, then uses an interrupt to notify kernel B, and finally it enters an infinite loop, which means that kernel A will never exit the program exception handling to prevent further exclusion of the command when it is executed again.

At step 104, the core B in the normal state receives information about the interrupt from core A, then awakens its process hidden from the user to eliminate the exceptional situation in the kernel, prepares for the search and finds which core is in the exceptional state, and prepares the recovery.

In step 105, core B uses a common control register to set core A to a reset state. Since in multi-core CPUs, when one core is set to a reset state, it does not execute any code, which means that it is in a “stop” state. As soon as this kernel is freed from the reset state, it will read and execute commands starting from a fixed start address, that is, it will perform a restart operation.

At step 106, core B notifies the system scheduling module. This module then transfers the tasks originally assigned to kernel A to another kernel in the normal state in accordance with its scheduling algorithm, thus ensuring the timeliness of the task.

At step 107, core B redistributes the resources originally occupied by core A in the system. These resources mainly include: task queue, stack space, interrupt, etc.

In step 108, core B uses the general control register of the CPU to free core A from the reset state, and core A starts again. Then, kernel B requests the value of the state of core A in the shared memory cell and waits until it changes to “waiting for recovery”.

In step 201, after kernel A is freed from the reset state, it starts reading and executing commands from the fixed start address of the CPU, and then starts again.

At step 202, the core And repeatedly performs the initialization operation. Since kernel A uses new resources, it can undoubtedly be able to restart successfully. After the launch is completed, core A changes its state in the shared memory cell to “waiting for recovery”, which indicates that core A has already started.

At step 203, core B determines that the state of core A changes to “waiting for recovery,” which indicates that core A has already started. Then, core B changes the state of core A in the shared memory cell to “normal”. Then, kernel B notifies the system planning module of the assignment of tasks to core A.

Elimination of the exception is completed.

Claims (5)

1. A way to eliminate an exception in one of the cores of a multicore system, including shared memory and a system scheduler module, which includes the following steps:
a) allocating a cell in shared memory, storing the value of each core and setting the initial value of the state of each core to “normal”;
b) when an exception occurs on one core, this core automatically executes the exception handling program, sets its state value to the value “exception” and notifies another selected core in the normal state, then the core in the exceptional state automatically enters an infinite loop;
c) the selected kernel in the normal state sets the kernel in the exceptional state to the reset state and notifies the system planning module, then this module will transfer the tasks originally assigned to the kernel in the exceptional state to any other kernel in the normal state, the selected core in the normal state will redistribute all resources occupied by the kernel in the exceptional state, and, finally, free the core in the exceptional state;
d) after the release from the reset, the kernel in an exceptional state will restart and set its status value to “waiting for recovery” after the start procedure is completed;
f) after the selected kernel in the normal state detects that the value of the kernel in the exceptional state is “waiting for recovery”, it will set the state value of this kernel to “normal” and notify the system planning module. 2. The method of eliminating the exceptional situation according to claim 1, wherein in step b, an interrupt is used to send an alert to the multi-core communication system. 3. The way to eliminate the exception according to claim 1, in which the system planning module will monitor the state of each core according to the corresponding state of the value stored in the memory cell, if this module detects that the state of one core is an exceptional state, the module will not be more Assign tasks to a kernel in exceptional condition. 4. The way to eliminate the exceptional situation according to claims 1, 2 or 3, in which in the case when the multicore system is a symmetric multicore system in step b, the selected core in the normal state can be any core in the normal state. 5. The way to eliminate the exceptional situation according to claims 1, 2, or 3, in which, in the case where the multicore system is a “master – slave” system, in step b, the selected core in the normal state is a kernel that has the “master” state.