53 results about "Os kernel" patented technology

The invention provides a Linux access control system based on attributes, and relates to the technical field of data access control. The system comprises a system monitor module, a data interaction module and a decision unit. The system monitor module collects attributes from a kernel and a user space and writes the attributes into the data interaction module; the access decision unit compiles the access control strategy into an eBPF program and mounts the eBPF program to a corresponding hook; executing the kernel to the hook, and triggering an eBPF program; an eBPF program queries a Flow rule; matching the attribute with the Flow rule, and if the matching is successful, executing a corresponding action; if all the Flow rules fail to match, executing a default action; the system can be expanded during operation, and can be loaded or unloaded based on dynamic loading characteristics and strategies of the eBPF program and the eBPF program during operation of the system, so that the problem that a kernel needs to be compiled in a traditional LSM scheme is solved; the method does not intrude the kernel, is completely based on an eBPF program, does not modify a kernel source code, and can guarantee the stability and compatibility.

The invention provides a kernel function trajectory-based deep learning inference service performance analysis method, which comprises the following steps of: based on service indexes and hardware theoretical computing power acquired from a production cluster, defining floating point operation times per request (FPR) index to quantify service resource efficiency, and identifying high FPR hotspot services; positioning a reasoning iteration candidate boundary based on a GPU kernel function trajectory, verifying iteration integrity through fingerprint matching and chi-square test, and calculating a second reasoning iteration number IIPS and a model reasoning efficiency MIE; aiming at calculation-intensive operators on the key path, combining a dynamic Roofline model to estimate an operator theoretical performance upper limit, and based on actual execution time, calculating efficiency and a BottleScore index to identify a key bottleneck operator; and outputting targeted optimization suggestions according to analysis results of service efficiency analysis, model efficiency analysis and operator efficiency analysis. According to the method, the inference behavior pattern can be automatically identified from massive kernel trajectories, and the efficiency loss of each level is quantified.

The invention discloses an operating system IO processing method supporting random IO merging, an operating system and a storage system, and belongs to the field of computer operating systems. The method comprises the steps that in a user mode, only one parameter copy is reserved for the same parameters of a plurality of random IO requests and recorded in a shared structure A, respectively recording different parameters of each random IO request into the shared structure body B to obtain a pseudo sequence request; the pseudo-sequence requests are issued to a system kernel, corresponding operations are executed on the pseudo-sequence requests in all layers of the system kernel in sequence, the batch operations are only executed once, and the non-batch operations are executed once for all the random IO requests respectively; the batch operation and the non-batch operation are operations executed for parameters in the shared structure body A and the shared structure body B respectively; and splitting the pseudo sequence request into random IO requests in an equipment driving layer of the kernel system, generating a corresponding equipment command, and submitting the equipment command to storage equipment. According to the invention, kernel overhead under the condition of high throughput can be relieved, so that the throughput is improved.

A processor includes a plurality of processing elements. The processor is configured to execute a work graph including a plurality of nodes representing kernels executable by one or more processing elements of the plurality of processing elements. A first processing element of the one or more of the processing elements associated with a first node of the plurality of nodes is configured to assign each logical division of a sparse input matrix to a bin of a plurality of bins. Responsive to a dispatch condition associated with a bin of the plurality of bins, the processor is configured to dispatch a workgroup to at least a second processing element associated with at least a second node of the plurality of nodes corresponding to the bin. The workgroup includes a plurality of work items based on one or more logical divisions of the sparse input matrix assigned to the bin.

The embodiment of the invention provides a kernel selection method and device during general matrix multiplication operation, equipment and a storage medium, and belongs to the technical field of computers. The method comprises the following steps: acquiring general matrix multiplication problem size data; preprocessing the general matrix multiplication problem size data to obtain general matrix multiplication problem size features; based on a pre-trained problem size encoder, mapping the general matrix multiplication problem size feature into a problem size embedded vector of a preset dimension; taking the problem size embedded vector as a query vector, and performing nearest neighbor vector search in a pre-configured vector database to obtain a kernel configuration feature with the highest similarity; and outputting the kernel configuration feature as a selection result. The method is used for improving the kernel selection efficiency and precision during the operation of the general matrix multiplication.

An apparatus for computing functions using polynomial-based approximation, comprising one or more processing circuitries configured for computing a polynomial-based approximant approximating a function by executing one or more iterations. Each iteration comprising computing the polynomial-based approximant using scaled fixed-point unit(s) according to a constructed set of coefficients, minimizing an approximation error of the computed polynomial-based approximant compared to the function while complying with one or more constraints selected from a group comprising at least: an accuracy, a compute graph size, a computation complexity, and a hardware utilization of the processing circuitry(s), adjusting one or more of the coefficients in case the approximation error is incompliant with the constraint(s) and initiating another iteration. The polynomial-based approximant and its adjusted set of coefficients for which the computed polynomial-based approximant complies with the constraint(s) may be output to one or more processing circuitries configured to approximate the function by computing the polynomial-based approximant.

This invention relates to a method for cross-version binary compatibility of Linux kernel modules, belonging to the field of computer operating system kernels. It includes the following steps: During the compilation phase, the compiler identifies specific keywords (__kabi_reloc_member, __kabi_check_member, __kabi_sizeof) and records instruction offsets, type identifiers, member variable name offsets, and relocation types, generating a .kabi_relocs segment. During the loading phase, the loader parses the .kabi_relocs segment, combines it with module BTF and kernel BTF information, and dynamically repairs instructions based on the relocation type. This invention achieves "compile once, run anywhere" binary compatibility for kernel modules, with the repair operation completed at load time, having no impact on runtime performance.

The invention relates to the technical field of kernel code generation, and discloses an automatic reconstruction kernel code generation method, which comprises the following steps: collecting example code data of a kernel part, and carrying out structured processing on example codes to obtain a logic transfer diagram of the example codes; constructing an example code feature extraction model to perform embedded representation on the logic transfer diagram of the example code, and converting an embedded representation result into structural features of the example code; semantic features of the function description text are extracted; and receiving the kernel demand description text by using the automatic reconstruction model of the depth kernel code obtained by optimization solution, and reconstructing the kernel code to generate the kernel code meeting the kernel demand description text. According to the method, feature extraction is carried out on structural features representing a logic sequence and a logic structure of an example code and semantic features representing semantic information of a function description text, and the example code similar to semantic of a kernel demand description text is selected to carry out code structure reconstruction and kernel code generation.

The invention relates to the technical field of software testing, and provides an automatic system call specification generation method and system for kernel fuzz testing. The method comprises the following steps: a static analysis stage: compiling a Linux kernel source code to be tested into an LLVM byte code, and performing deep static analysis on the LLVM byte code to identify and reconstruct an interface and a parameter type called by a system; a symbol execution and specification generation stage: adopting a constraint extraction and solution algorithm to extract constraints under different paths, carrying out constraint solution, and generating an initialized calling specification template in combination with an interface called by the system, the parameter type and the constraints; a standard test and verification stage: constructing an evaluation system of a multi-dimensional index to verify the initialized calling standard template and generate corresponding error information; in the large language model auxiliary correction stage, the initialized calling standard template and the error information serve as input, and a high-quality calling standard template is obtained after a large model iteration repair algorithm is conducted.

The invention belongs to the technical field of computer software testing, and particularly relates to a Linux kernel vulnerability mining method based on diversity guidance. According to the method, firstly, collected PoCs are expressed by using a customized abstract syntax tree, clustering is carried out on the PoCs based on a Louvain community discovery algorithm, an initial diversity seed bank is constructed, and seeds are divided into a plurality of communities with different functions; in order to quantify seed diversity, a community prevalence rate index (CPR) is introduced; designing a double-layer multi-arm tiger machine scheduling framework based on the CPR, wherein the framework is used for efficiently allocating variable resources between communities and in the communities; a CPR-guided seed variation strategy is adopted to preferentially carry out rapid variation and expansion on high-diversity seeds, so that the coverage speed and efficiency of vulnerability triggering are improved. Experimental results show that compared with a current most advanced kernel fuzzy test tool, the method has the advantages that the code coverage rate is averagely increased by 17.4%, and the vulnerability discovery number is averagely increased by 9.1 times.

The invention discloses a code generation method and electronic equipment, and relates to the technical field of artificial intelligence. The method comprises the following steps: generating a knowledge base according to an operation form of performing data operation on task data by an operator and an implementation demand; according to the hardware architecture data of the operator running target hardware and the knowledge base, generating a cue word of a kernel code supporting execution of a corresponding task on the target hardware; inputting the kernel code generation cue word into a code generation language model, and obtaining a kernel code generation result output by the language model; and if the kernel code generation result does not pass the test verification, adding error information in the test verification process to a kernel code generation prompt word, and re-inputting the kernel code generation prompt word to the language model until the kernel code generation result passes the test verification. According to the method and the device, the problem that hardware adaptation is difficult in the kernel code optimization process in the prior art can be solved, the kernel codes are automatically generated in the whole process, and the kernel code generation quality and the kernel code generation efficiency are improved.

The present disclosure provides a model compilation security evaluation method and system based on structural detection and differential analysis, relating to the technical field of compilation security evaluation, comprising: calling a compiler for compilation to generate intermediate representations and compilation products in each stage; performing single-file static detection on the intermediate representation graph file to identify neural network backdoor patterns embedded in the form of mathematical calculation structures in the graph; reading the intermediate representation graph file before and after operator fusion, performing differential analysis on the intermediate representation graph file before and after fusion, and identifying abnormal operators or data streams injected in the fusion stage; performing differential analysis on the kernel source code file and the machine code disassembly file to identify malicious tampering injected in the code generation or binary compilation stage; and synthesizing the detection results of multiple identification processes to perform final security evaluation on the model compilation process and generate a structured evaluation report. The present disclosure improves the identification capability of abnormal structures or potential backdoor logic hidden in the model compilation stage.

This application discloses a kernel probe generation method, apparatus, computer device, and storage medium, relating to the field of operating system kernel monitoring technology. The method includes: responding to a kernel probe generation request, obtaining a target metadata file corresponding to the current operating system kernel; determining a target declarative policy file; determining a target program logic file corresponding to the target declarative policy file; and generating probe code based on the target program logic file and the target metadata file to obtain a kernel probe program. This application, by combining the target metadata file and the target declarative policy file, can dynamically generate probe program source code adapted to the current kernel version. This source code generation function decouples the kernel probe development logic from underlying kernel knowledge, eliminating the need for developers to deeply understand the kernel's internal data structures and version differences, thus lowering the development threshold. It supports not only specific operating systems or probe types but also the general Linux kernel and various probe types.

The invention discloses a server network monitoring method and device, and relates to the technical field of server monitoring, and the method comprises the steps: obtaining the function index information of a plurality of target kernel network functions in a BTF type function database; the BTF type function database comprises a plurality of kernel network functions and the function index information of each kernel network function; the target kernel network function is a kernel network function comprising a target parameter in the BTF type function database; for each target kernel network function, determining a position sequence number of a target parameter in the target kernel network function in a parameter list according to the function index information; according to the position serial number, matching a corresponding kernel probe program from a plurality of preset kernel probe programs, and mounting the kernel probe program to a corresponding target kernel network function; and when the target kernel network function is called, collecting kernel network event data through a kernel probe program. According to the method and the device, the monitoring efficiency of the Linux kernel network can be improved.

The application discloses a kind of based on kernel event-driven Linux system fault diagnosis method and system;Its operation steps: real-time capture hard exception, soft exception and resource exception event;Dynamic monitoring is carried out to CPU, memory, storage I / O and network resource use trend and mutation behavior;Complete call stack, execution parameter, lock contention state and connection metadata are formed structured fault context information when abnormal moment is comprehensively collected, CPU register state is urgently frozen to firmware reserved memory in interrupt context respectively, task structure body and kernel stack information are collected in software interrupt context, cross-CPU core state snapshot is completed in high-priority kernel thread coordination;According to system load, direct memory access transmission, atomic log addition or bare device block write mode is selected adaptively, and collected diagnostic data is written and saved;System recovery agent automatically reorganizes diagnostic data, and generates standardized diagnostic package.The application realizes the deep observability of Linux system while ensuring very low running overhead, and significantly improves fault positioning efficiency.

The invention discloses a mapping method for a computational graph and hardware fusion kernel, which comprises the following steps of: S1, analyzing a fusion kernel and a corresponding sub-graph structure in hardware, and establishing a fusion mode based on a fusion rule; s2, loading an initial calculation graph exported by the analytic reasoning framework; s3, matching each candidate node and each sub-graph in the calculation graph with each fusion mode; and S4, repeatedly executing the step S3 until the complete calculation graph is traversed and no new matching occurs, and generating an optimized calculation graph. The invention further provides a system for implementing the method, and the system has wide application value.

The invention belongs to the technical field of operating system kernel vulnerability analysis, and particularly relates to an instruction-level kernel vulnerability triggering process tracking method based on task selectivity. The method comprises the steps that all kernel tasks triggered by a vulnerability concept verification program are recognized in the execution process of the vulnerability concept verification program, and a range is defined for subsequent selective execution; in the execution process of the vulnerability concept verification program, the instruction executed by the kernel is selectively tracked according to the defined execution range, and the kernel instruction level tracking efficiency is improved; and finally, a kernel execution record is stored in a high-density binary storage format, so that the storage space is greatly saved under the condition of ensuring complete data storage. According to the method, through selective kernel execution tracking, the efficiency of instruction-level kernel execution tracking is effectively improved, the storage space required by kernel execution tracks is reduced, and an efficient, accurate and complete vulnerability triggering process tracking tool is provided for various kernel vulnerability dynamic analysis tasks.

The application provides a target detection system and method based on a Linux kernel V4L2 driver, the system comprising a hardware layer, a kernel space performing target detection preprocessing, and a user space for image display; the kernel space comprising: a V4L2 driving module acquiring original video frames, using a lightweight CNN model as an AI preprocessing model, extracting and directly outputting a feature vector; an asynchronous inference module using an asynchronous inference mechanism to run an AI inference task of the AI preprocessing model, determining an inference result of the feature vector, and writing the inference result into an extension field of a V4L2 buffer area so that the user space reads the inference result for image display. The application can avoid the redundancy of copying frame data from the kernel state to the user state for reprocessing, avoid the problems of delay and high CPU resource occupancy rate caused by user space AI processing, reduce data transmission delay and CPU occupancy rate, and improve system stability and multitasking processing capacity.

The application discloses an operating system fault rehearsal method and device based on a kernel debugging probe, equipment and a medium, relates to the field of system debugging, and comprises the following steps: updating a preset fusion weight matrix based on the probe triggering condition of a preset kernel debugging probe in a target operating system to obtain an updated fusion weight matrix; constructing a system feature vector according to the current system data of the target operating system, and calculating a system feature value according to the system feature vector and the updated fusion weight matrix; determining whether to perform fault rehearsal on the target operating system based on the system feature value, and if so, performing fault node positioning based on the system feature vector to determine a target fault node; and running a plurality of repair schemes corresponding to the target fault node in a preset environment, and selecting a target repair scheme from the plurality of repair schemes based on the obtained running result. Thus, the dynamic positioning and analysis of kernel faults can be realized through the kernel probe.

The invention belongs to the related technical field of hardware acceleration, and particularly relates to a hardware acceleration system and method oriented to large model reasoning, and the system comprises an FPGA reasoning acceleration module; according to calculation dimensions of different stages in the reasoning process, a linear calculation kernel, a multi-head attention calculation kernel, a feedforward neural network layer calculation kernel and a quantization normalization and residual connection kernel are divided; the four cores are distributed in different SLR regions of the FPGA, and the whole of each core is located in the same SLR region; hardware-level FIFO queues are integrated at data transmission interfaces among the cores; blocking type data flow control logic is adopted, each kernel autonomously starts or pauses calculation operation of the kernel only according to data availability of an input queue and space availability of an output queue of the kernel, a calculation result is directly pushed to an input queue of a downstream kernel, and a self-triggering execution chain based on data flow is formed. According to the method, the problems of relatively low reasoning efficiency and high resource consumption of the edge scene under the word element sequence can be solved.

The invention relates to a process wake-up relation tracking system and method based on an eBPF, and belongs to the technical field of computer system performance analysis, the process wake-up relation tracking system comprises a kernel mode BPF program and a user mode processing program, the kernel mode BPF program is deployed in a Linux kernel and is used for capturing a process wake-up event and recording a wake-up relation; the user mode processing program runs in a user space and is used for reading and analyzing data collected by the kernel mode BPF program; according to the method, the BPF program is deployed in the Linux kernel, the wake-up event between the processes is captured in real time, the process wake-up tree and the wake-up tree are constructed and maintained, and multi-level tracking of the direct and indirect wake-up relation of the target process is achieved. According to the method, detailed information, including the process ID, the process name, the timestamp and the priority, of the wake-up event can be accurately recorded, the tracking data is processed and displayed through the user mode program, and a key basis is provided for system performance analysis and process scheduling optimization.

The application provides a lightweight universal compiling method, system and device. The compiling method comprises the following steps: obtaining kernel code, dynamic library loading code, C function library code, limited calling interface library function and calling agent code of a microkernel operating system; obtaining a compiler of a processor; compiling the kernel code, the dynamic library loading code, the C function library code and the limited calling interface library function by using the compiler to obtain a compatible kernel, a compatible dynamic library loader, a compatible C function library and a compatible limited calling interface library; modifying and / or supplementing the calling agent code according to the architecture of the processor, and compiling the modified and / or supplemented calling agent code by using the compiler to obtain a compatible calling agent library. The compiling system comprises various functional modules corresponding to the above steps. The compiling device is an electronic device capable of implementing the above-mentioned compiling method. The application can realize the decoupling of the microkernel operating system kernel and the instruction set and guarantee the testability of the application program.

The invention provides a model compilation safety assessment method and system based on structure detection and differential analysis, and relates to the technical field of compilation safety assessment, and the method comprises the steps: calling a compiler for compiling, and generating an intermediate representation of each stage and a compilation product; performing single-file static detection on the intermediate representation graph file, and identifying a neural network backdoor mode embedded in the graph in a mathematical calculation structure form; reading an intermediate representation graph file before and after operator fusion, carrying out differential analysis on the intermediate representation graph file before and after the operator fusion, and identifying an abnormal operator or a data stream injected in a fusion stage; performing differential analysis on the kernel source code file and the machine code disassembling file, and identifying malicious tampering injected in a code generation or binary compiling stage; and synthesizing the detection results of the plurality of identification processes, performing final security assessment on the model compiling process, and generating a structured assessment report. According to the method and the device, the recognition capability of an abnormal structure or potential backdoor logic hidden in a model compiling stage is improved.

The invention relates to the technical field of fuzz testing, and provides a kernel fuzz testing method guided by a characteristic interface and related equipment, and the method comprises the steps: obtaining the specific operation interface information of a target release version kernel based on the target release version kernel and a main line kernel corresponding to the target release version kernel; generating an interface test case of the target release version kernel according to the specific operation interface information by utilizing a large language model; obtaining a plurality of general test cases, and determining a priority value of each general test case based on the interface test case; and performing fuzz testing on the target release version kernel by using the priority values of all the general test cases to obtain a fuzz testing result of the target release version kernel. According to the method provided by the invention, the effectiveness of the fuzz test on the release version kernel and the main line kernel corresponding to the release version kernel can be improved.

The invention belongs to the technical field of GPU performance prediction. The invention provides a kernel-level execution time prediction method for a GPU deep learning task. According to the embodiment of the invention, the cloud platform can quickly and accurately predict the execution time of any deep learning task on different GPU models before the GPU resources are allocated, and meanwhile, the dependence on large-scale data collection and a complex model retraining process are avoided. The kernel-level execution time prediction system adopts a training-free system design, kernel-level execution time prediction is cooperatively realized through a core component consisting of an analyzer (comprising a client analyzer and a server analyzer), a predictor, a kernel mapping table and a kernel cache, the analyzer is responsible for collecting general kernel information across clients and servers, and the predictor is responsible for predicting kernel execution time. The predictor generates execution time on the target GPU based on kernel information, the kernel mapping table is used for aligning kernel sequences under different GPU architectures, and the kernel cache continuously records and reuses observed kernel execution features.

The application provides a multi-kernel geometric object and algorithm adaptive mapping method and system, which comprises the following steps: facing a CAD platform layer capable of supporting multiple geometric kernels, a developer selects a target geometric kernel to be bound from candidate geometric kernels before system compilation or construction, and determines the corresponding relationship between the topological object, geometric object, attribute data, precision data and native algorithm function of the selected geometric kernel and the platform unified topological expression and unified algorithm interface according to the pre-defined mapping rule corresponding to the selected geometric kernel; in the compilation or construction stage, the mapping rule is solidified into the corresponding kernel adaptive interface implementation, object construction rule and algorithm encapsulation logic; in the system running period, the upper-layer application performs modeling operation through the unified geometric object model and unified algorithm interface, and the lower layer fixedly calls the geometric kernel native object and native function which have been bound in the compilation period.

This application provides a QoS control method and system based on IO demand prediction. The method includes: capturing the IO system call trajectory of a target AI task and extracting an enhanced IO feature vector, where the enhanced IO feature vector is a feature vector containing system call features and file metadata features; matching a pre-trained enhanced IO fingerprint library to identify the current execution stage; predicting the IO demand of the next stage based on a Hidden Markov Model (HMM), where HMM is a probabilistic model used for sequence state prediction; and directly calling the kernel's native block device layer QoS interface to update the token bucket parameters of the target AI task based on the prediction results, where the block device layer QoS interface refers to the token bucket parameter update interface provided by the Linux kernel's rq-qos module. This application achieves non-intrusive prediction of AI task IO demands and ultra-low latency dynamic control of block device layer bandwidth without modifying application code or kernel code.

The invention discloses an AI accelerator optimization method based on a multipath kernel software stack and a system thereof, which are characterized in that a zero-copy, automatic reverse address mapping and dynamic load balancing method is adopted, high-compatibility accelerator access support is realized, efficient reasoning is realized through a double-path architecture, and the stability of the system is improved. The slow path calls a user mode accelerator library through a user mode daemon process to provide wide compatibility, the fast path realizes direct access of a kernel to an accelerator through a hijacking working environment, and zero-copy data transmission, automatic reverse address mapping, dynamic load balancing and a self-adaptive water level mechanism are introduced to further optimize memory management and task scheduling. Compared with the prior art, the method has the advantages that inherent delay and end-to-end delay of kernel machine learning reasoning are reduced, kernel-user space communication overhead is eliminated, system performance is remarkably improved, the problems of high reasoning delay and poor compatibility caused by the fact that an operating system kernel cannot directly access a hardware accelerator are solved, and the method has good application prospects.