WO2017107163A1

WO2017107163A1 - Memory management method and system based on heterogeneous hybrid memory

Info

Publication number: WO2017107163A1
Application number: PCT/CN2015/098817
Authority: WO
Inventors: 马先明; 庞观士; 徐成泽; 沈航; 陈志列
Original assignee: 研祥智能科技股份有限公司
Priority date: 2015-12-25
Filing date: 2015-12-25
Publication date: 2017-06-29

Abstract

A memory management method based on a heterogeneous hybrid memory. The method comprises: receiving a memory application request (S110); detecting whether an NVM mark exists in the memory application request (S120); if the NVM mark exists in the memory application request, judging whether the memory application request meets the permission requirements (S130), if so, distributing an NVM according to the memory application request (S140), and if not, ending; and, if the NVM mark does not exist in the memory application request, distributing a DRAM according to the memory application request (S150). According to the method, by detecting whether an NVM mark exists in a memory application request, and judging whether the memory application request meets the permission requirements, whether the DRAM or the NVM is distributed is determined according to detection and judgment results, so that the NVM and the DRAM can be managed simultaneously. In addition, also provided is a memory management system based on a heterogeneous hybrid memory.

Description

Memory management method and system based on heterogeneous mixed memory

Technical field

The present invention relates to the field of computer technologies, and in particular, to a memory management method and system based on heterogeneous mixed memory.

Background technique

With the development of emerging non-volatile memory (Non-Volatile Memory, NVM) technology represented by resistor memory, ferroelectric memory, phase change memory, etc., the development of memory technology is advanced, and it is a new type of memory and storage system. The structure has laid a good foundation. By combining the new NVM with Dynamic Random Access Memory (DRAM), a hybrid memory architecture is built to form heterogeneous mixed memory. Heterogeneous hybrid memory has the characteristics of NVM and DRAM at the same time. It not only has the function of conventional memory, but the data stored in DRAM will disappear immediately after power off, which is volatile. At the same time, it can also play the NVM after power failure. The ability to save data and take full advantage of the non-volatile nature of NVM is a new type of memory that is used in combination.

Heterogeneous hybrid memory adopts DDR3 (Double Data Rate, double-rate synchronous dynamic random access memory, third generation) interface memory module design to meet the conventional memory interface of existing industrial control equipment, without introducing new industrial control equipment or adding New auxiliary equipment has become a research hotspot.

The application field of heterogeneous mixed memory is mainly the industrial control field that requires big data and large memory computing. The Linux operating system is widely used in the industrial control field because of its open source code, powerful function and easy migration and optimization. Therefore, the memory management of heterogeneous mixed memory is implemented on the Linux operating system. The promotion of memory is of great significance.

As a new type of memory, heterogeneous mixed memory integrates NVM and DRAM. The conventional memory management method is only the management of DRAM memory, which cannot satisfy the management and operation of NVM memory. Therefore, it will not be used for heterogeneous mixed memory. It is applicable, so it is necessary to provide a memory management method that can be applied to heterogeneous mixed memory, and can simultaneously manage NVM memory and DRAM memory.

Summary of the invention

Based on this, it is necessary to provide a memory management method and system based on heterogeneous mixed memory for the above technical problems, and can simultaneously manage NVM memory and DRAM memory.

A memory management method based on heterogeneous mixed memory, the method comprising:

Receiving a memory application request;

Detect whether there is an NVM flag in the memory request request;

If the NVM flag exists in the memory application request, determining whether the memory application request meets a permission requirement, and if yes, allocating NVM memory according to the memory application request, and if not, ending;

If the NVM flag does not exist in the memory request request, the DRAM memory is allocated according to the memory request.

In one embodiment, the memory request request includes a memory length of the application; before the step of determining whether the memory application request meets the permission requirement, the method further includes:

Determining whether the memory length of the application exceeds the total length of the NVM memory, and if not, entering the step of determining whether the memory application request meets the permission requirement; if yes, ending.

In one embodiment, the memory request request includes a memory length of the application; the root The steps for allocating NVM memory according to the request for the application include:

Applying and initializing a virtual address space of the same size according to the memory length of the application, where the virtual address space includes a first address and a tail address of the virtual address space;

Converting the first address of the virtual address space into a physical address, and applying NVM memory of the same size in the NVM according to the physical address and the requested memory length.

In one of these embodiments, the virtual address space is located in an indirect mapped address space in the kernel space of Linux.

In one of the embodiments, the physical address of the NVM memory is pre-allocated in the ZONE_HIGHMEN area in the Linux kernel memory management area, and is connected to the ZONE_NORMAL area in the Linux kernel memory management area.

A memory management system based on heterogeneous mixed memory, the system comprising:

a receiving module, configured to receive a memory application request;

a detecting module, configured to detect whether an NVM flag exists in the memory application request;

a first determining module, configured to determine, if the NVM flag exists in the memory application request, whether the memory application request meets a permission requirement, and if yes, allocate NVM memory according to the memory application request, and if not, not process ;

An application module, configured to allocate NVM memory according to the memory application request if the memory application request meets a permission requirement;

The application module is further configured to allocate DRAM memory according to the memory application request if the NVM flag does not exist in the memory application request or does not meet the permission requirement.

In one embodiment, the memory request request includes a memory length of the application, and the system further includes:

The second judging module is configured to determine whether the memory length of the application exceeds the total length of the NVM memory, and if not, notify the first judging module whether the memory application request meets the permission requirement, and if not, not deal with.

In one embodiment, the memory request request includes a memory length of the application, and the application module includes:

a virtual address space application unit, configured to apply and initialize a virtual address space of the same size according to a memory length of the application, where the virtual address space includes a first address and a tail address of the virtual address space;

An address conversion unit, configured to convert a first address of the virtual address space into a physical address;

The memory application unit is configured to apply for the same size of the NVM memory in the NVM according to the physical address and the requested memory length.

The above heterogeneous mixed memory based memory management method and system, by detecting whether there is an NVM flag in the memory application request, and further determining whether the memory application request meets the permission requirement when the NVM flag exists in the memory application request, and according to the detection and The result of the judgment is to allocate DRAM memory or NVM memory. This memory management method and system can be compatible with the application and allocation of conventional DRAM memory, and can introduce new NVM memory application and allocation, so that NVM memory and DRAM memory can be managed simultaneously to meet the use of heterogeneous mixed memory.

DRAWINGS

1 is an interface diagram of a heterogeneous mixed memory in one embodiment;

2 is a flow chart of a memory management method based on heterogeneous mixed memory in one embodiment;

3 is a flow chart of a memory management method based on heterogeneous mixed memory in another embodiment;

4 is a flow chart of allocating NVM memory according to an application request in an embodiment;

5 is a distribution diagram of a virtual address space in a 64-bit Linux system in an embodiment;

6 is a physical address space distribution diagram of a heterogeneous mixed memory in an embodiment;

7 is a structural block diagram of a memory management system based on heterogeneous mixed memory in one embodiment;

8 is a structural block diagram of a memory management system based on heterogeneous mixed memory in another embodiment;

Figure 9 is a block diagram showing the structure of an application module in one embodiment.

detailed description

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The heterogeneous hybrid memory adopts a physical connection method consistent with the DDR3 (Double Data Rate, double-rate synchronous dynamic random access memory, third generation) interface circuit. As shown in Figure 1, it is an interface diagram of heterogeneous mixed memory. The memory module is inserted into the commonly used DDR3 interface circuit through the JEDEC (Joint Electron Device Engineering Council) standard interface connector, and the address (Address, Addr) and command transmitted by the operating system to the heterogeneous mixed memory ( Information such as Command, Cmd, and Data can be input to the controller through the interface circuit. The Controller controls the DRAM memory and the NVM memory through the received data and internal logic. among them The controller can be an MCU (Micro Control Unit), an FPGA (Field-Programmable Gate Array), a CPLD (Complex Programmable Logic Device), or an ARM (Advanced RISC Machine, Advanced). The reduced instruction set machine) is implemented by a processor, and provides corresponding read and write control and timing operations according to the respective characteristics of the DRAM and the NVM. The present invention is a memory management method and system based on heterogeneous mixed memory implemented in an interface environment as described above.

In an embodiment, as shown in FIG. 2, a memory management method based on heterogeneous mixed memory is provided, and the method includes:

Step S110, receiving a memory application request.

In this embodiment, the operating system driver, the library function, and the application program may need to apply for memory. When the memory is applied, the memory application request is sent to the operating system kernel, and the operating system kernel allocates the memory according to the received memory application request.

Step S120, detecting whether there is an NVM flag in the memory application request, if yes, proceeding to step S130, otherwise proceeding to step S150.

In this embodiment, there may be an NVM flag in the memory request, such as the last bit stored in the memory request. If it is detected that there is an NVM flag in the memory request request, it indicates that NVM memory needs to be applied. If not, it indicates that you need to apply for DRAM memory.

Step S130, determining whether the memory application request meets the permission requirement, if yes, proceeding to step S140, otherwise ending.

User permissions are an aspect of Linux security, and different users have different permissions. The root user can have any files and commands in the operating system and has the highest authority. You can specify which users have permission to allocate NVM memory as needed. In this embodiment, the NVM memory in the heterogeneous mixed memory The application only allows the root user to operate, and other users do not have permission to apply for NVM memory. If the memory application request is sent by the root user, it is determined to meet the permission requirement, otherwise the permission requirement is not met.

Step S140, allocating NVM memory according to the memory application request.

In this embodiment, the NVM memory and the DRAM memory can be uniformly addressed in advance, and paging is performed in units of physical pages. For example, the size of each physical page defaults to 4 KB. The NVM memory and DRAM memory are divided into different Linux kernel memory management areas. The Linux kernel memory management is divided into ZONE_DMA, ZONE_NORMAL and ZONE_HIGHMEM, and the NVM memory can be divided into any memory management area. The request may include information such as the length of the application's memory, the type of memory management area requested, and the memory continuity requirements of the application. The memory length of the application is used to determine the size of the memory to be allocated. The type of the memory management area to be used is used to determine which of the above memory management areas is allocated NVM memory, and the requested memory continuity requirement is used to determine whether to allocate contiguous memory. The Linux kernel can implement memory allocation based on this information.

Step S150, allocating DRAM memory according to the memory application request.

As described above, the NVM memory and the DRAM memory are collectively addressed in advance, and paging is performed in units of physical pages. For example, the size of each physical page defaults to 4 KB. The NVM memory and DRAM memory are divided into different Linux kernel memory management areas. The Linux kernel memory management is divided into ZONE_DMA, ZONE_NORMAL and ZONE_HIGHMEM, and the DRAM memory can be divided into any memory management area. The request may include information such as the length of the application's memory, the type of memory management area requested, and the memory continuity requirements of the application. The memory length of the application is used to determine the size of the memory to be allocated. The type of memory management area used is used to determine which of the above memory management areas is allocated DRAM memory. The requested memory continuity requirement is used to determine whether to allocate contiguous memory. The Linux kernel can implement memory allocation based on this information.

In this embodiment, by detecting whether the NVM flag exists in the memory application request, and when the NVM flag exists in the memory application request, further determining whether the memory application request meets the permission requirement, and selecting to allocate the DRAM memory according to the result of the detection and the judgment NVM memory. This memory management method and system can be compatible with the application and allocation of conventional DRAM memory, and can introduce new NVM memory application and allocation, so that NVM memory and DRAM memory can be managed simultaneously to meet the use of heterogeneous mixed memory.

In addition, before the NVM memory is allocated, it is determined whether the memory application request meets the permission requirement, and only if the memory application request meets the permission requirement, for example, only the root user can request to allocate the NVM memory, so that only the user with the highest authority can be stored in the In the non-volatile storage space, unnecessary waste of storage space is avoided, and security is also ensured.

In another embodiment, as shown in FIG. 3, a memory management method based on heterogeneous mixed memory is provided, and the method includes:

Step S210, receiving a memory application request.

Step S220, detecting whether there is an NVM flag in the memory application request, if yes, proceeding to step S230, otherwise proceeding to step S260.

In step S230, it is determined whether the requested memory length exceeds the total length of the NVM memory. If not, the process proceeds to step S240; if not, the process ends.

In this embodiment, the memory application request includes the requested memory length. If the requested memory length exceeds the total length of the NVM memory, the NVM memory cannot satisfy the memory application request, and the application can be directly ended, and a flag indicating that the allocation fails is returned. .

Step S240, determining whether the memory application request meets the permission requirement, if yes, proceeding to step S250, otherwise ending.

Step S250, allocating NVM memory according to the memory application request.

Step S260, allocating DRAM memory according to the memory application request.

In this embodiment, when the requested memory length exceeds the maximum allowed by the total capacity of the NVM memory, it indicates that the NVM memory cannot meet its storage requirement and can be directly terminated. In this way, the efficiency of NVM memory allocation can be improved.

In one embodiment, the memory request request includes the memory length of the application. As shown in FIG. 4, the steps of allocating NVM memory according to the request for application include:

Step S251, applying and initializing a virtual address space of the same size according to the requested memory length, where the virtual address space includes a first address and a tail address of the virtual address space.

Step S252, converting the first address of the virtual address space into a physical address.

Step S253, applying NVM memory of the same size in the NVM according to the physical address and the requested memory length.

The application field of heterogeneous mixed memory is mainly for the application of big data and large memory computing in industrial control field. Its addressing space is large. In order to improve the reliability of big data access, heterogeneous mixed memory is often used in 64-bit processors. On the system. Therefore, memory management of heterogeneous mixed memory on a 64-bit Linux system is of great significance for the promotion of heterogeneous mixed memory. The principle of the memory management method for the heterogeneous mixed memory provided by the embodiment of the present invention is described in detail below by taking a 64-bit Linux system as an example.

As shown in Figure 5, it is a distribution map of 64-bit Linux virtual address space. In 64-bit Linux systems, the virtual address space includes 128T user space and 128T kernel space, which is a huge gap between user space and kernel space. Among them, from 0x0000, 0000, 0000, 0000 to 0x0000, 7FFFF, FFFF, F000, a total of 128T virtual address range is user space, 0xFFFF, 8000, 0000, 0000 or more virtual address space a total of 128T for kernel space, and slave address 0x0000, 7FFF, FFFF, F000 to 0xFFFF, 8000, 0000, 0000 is a huge void, reserved for future expansion.

In the kernel space, the 64T virtual address space from address 0xFFFF, 8800, 0000, 0000 to 0xFFFF, C7FF, FFFF, FFFF can be directly mapped to physical memory, that is, directly mapped to DRAM memory in heterogeneous mixed memory. This is consistent with the distribution of the traditional Linux kernel virtual address space.

Further, in the kernel space, the 32T address space between the addresses 0xFFFF, C900, 0000, 0000 to 0xFFFF, C8FF, FFFF, FFFF is an indirect mapped address space in the kernel space of Linux. The so-called indirect mapping address space means that the physical memory is not directly mapped. The address space of this part is used by the preset function, and the virtual address space is requested by a preset function. Since the virtual address space is not directly mapped to the physical memory, the applied virtual address space cannot obtain the physical address through a simple linear relationship, but needs to be converted into a physical address according to the mapping relationship, thereby allocating the memory according to the converted physical address. .

The default functions include vmalloc function, ioremap function and nvmmalloc function. The vmalloc function and ioremap are functions related to DRAM memory. You can call vmalloc to allocate DRAM memory, and call ioremap to input I/O (input/output, input/output port) interface. The DRAM memory is mapped to the virtual address space of the kernel. The nvmmalloc function is newly added to allocate NVM memory. The virtual address space allocated by these three functions is dynamic and random in the above 32T space, and there is no strict division. The three of them use the address together until the address is exhausted.

In this embodiment, the memory application request includes the memory length of the application, for example, the size parameter is used to identify the memory length of the application, and is used as an input parameter of the nvmmalloc function to allocate NVM memory through the nvmmalloc function. When the nvmmalloc function is called, it is virtualized according to the input parameter size. The space of the same size is applied in the address space and initialized, and the value in the requested space is cleared by the initialization. The nvmmalloc function returns the first address of the requested virtual address space. Specifically, the virtual address space applied by the nvmmalloc function is located in the indirect mapping address space in the kernel space of the above Linux system.

In addition, you can define other parameters of the nvmmalloc function. For example, the definition of flags indicates whether a virtual address space has been allocated. When a virtual address space has been allocated, it is marked to prevent duplicate allocation. It is also possible to define nvmflags to indicate that the allocated virtual address space is mapped to the NVM memory.

At the same time, when the nvmmalloc function is called, the controller in the heterogeneous mixed memory is triggered, and the controller converts the first address of the virtual address space into a physical address. Specifically, the Linux kernel sends the first address of the virtual address space to the controller through the interface circuit, where the mapping relationship between the virtual address and the physical address is stored, and the virtual address can be converted into a physical address, and according to the physical address and The requested memory length is the same size NVM memory in the NVM.

In one embodiment, the physical address of the NVM memory is pre-allocated in the ZONE_HIGHMEN area of the Linux kernel memory management area, and is connected to the ZONE_NORMAL area in the Linux kernel memory management area.

Specifically, as shown in FIG. 6, the NVM memory and the DRAM memory are uniformly addressed in advance, and all physical memory is paged by a fixed size, and all physical pages are divided into three types of memory management areas. The range of the ZONE_DMA area is 0 to 16 M. The physical page of this area is dedicated to DMA (Direct Memory Access) of the I/O device. The reason why the physical page of the DMA needs to be managed separately is because the DMA directly accesses the memory using the physical address and needs a continuous buffer. Therefore, in order to provide a physically contiguous buffer, a region must be specifically allocated from the physical address space for DMA. . The range of the ZONE_NORMAL area is 16M to 60T, the area of the area The page is directly mapped to the virtual address space and is DRAM memory. The space above 60T is the ZONE_HIGHMEM area, and the physical page and virtual address space of this area are indirectly mapped. The NVM storage medium that can be used has a low access speed and is inefficient. In this embodiment, the NVM memory and the virtual address space are not directly mapped, but the physical address of the NVM memory is pre-allocated in the ZONE_HIGHMEN area. And connected to the ZONE_NORMAL area. In addition, the size of the NVM memory affects the size of the ZONE_NORMAL area. The total length of the ZONE_DMA area, the ZONE_NORMAL area, and the NVM memory is 64T. For example, in this embodiment, the NVM memory is 4T, and the tail address of the ZONE_NORMAL area is 60T. The space from 60T to 64T is NVM memory indirectly mapped to the virtual address space. The space from 64T to 256T is consistent with the traditional high-end memory, which can be allocated and used in the traditional way.

In this embodiment, the NVM memory and the DRAM memory are uniformly and reasonably distributed in the 64-bit Linux system, and the corresponding physical address is efficiently found by the controller through the virtual address, and the NVM memory and the DRAM memory can be quickly allocated.

As shown in FIG. 7, in one embodiment, a memory management system based on heterogeneous mixed memory is provided, including:

The receiving module 310 is configured to receive a memory application request.

The detecting module 320 is configured to detect whether an NVM flag exists in the memory application request.

The first determining module 330 is configured to determine, if the NVM flag exists in the memory application request, whether the memory application request meets the permission requirement, and if yes, allocate the NVM memory according to the memory application request, and if not, not process.

The application module 340 is configured to request according to the memory request if the memory application request meets the permission requirement Allocating NVM memory; also used to allocate DRAM memory according to the memory request request if the NVM flag does not exist in the memory request request or does not meet the permission requirement.

The memory request request may include information such as the requested memory length, the type of memory management area requested, and the memory continuity requirement of the application.

As shown in FIG. 8, in another embodiment, a memory management system based on a heterogeneous mixed memory is provided. The system shown in FIG. 7 further includes:

The second determining module 350 is configured to determine whether the requested memory length exceeds the total length of the NVM memory. If not, the first determining module 330 is notified whether the memory request request meets the permission requirement, and if not, the processing is not performed.

In one embodiment, the memory application request includes the memory length of the application. As shown in FIG. 9, the application module 340 includes:

The virtual address space application unit 341 is configured to apply for and initialize the virtual address space of the same size according to the requested memory length, where the virtual address space includes a first address and a tail address of the virtual address space;

An address conversion unit 342, configured to convert a first address of the virtual address space into a physical address;

The memory application unit 343 is configured to apply for the same size of NVM memory in the NVM according to the physical address and the requested memory length.

In one embodiment, the virtual address space of the above system is located in an indirect mapped address space in the kernel space of Linux.

In one embodiment, the physical address of the NVM memory of the above system is pre-allocated in the ZONE_HIGHMEN area of the Linux kernel memory management area, and is connected to the ZONE_NORMAL area in the Linux kernel memory management area.

The above-mentioned embodiments are merely illustrative of several embodiments of the present invention, and the description thereof is more specific and detailed, but is not to be construed as limiting the scope of the invention. It should be noted that a number of variations and modifications may be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, the scope of the invention should be determined by the appended claims.

Claims

A memory management method based on heterogeneous mixed memory, the method comprising:

Receiving a memory application request;

Detect whether there is an NVM flag in the memory request request;

If the NVM flag exists in the memory application request, determining whether the memory application request meets a permission requirement, and if yes, allocating NVM memory according to the memory application request, and if not, ending;

If the NVM flag does not exist in the memory request request, the DRAM memory is allocated according to the memory request.
The method according to claim 1, wherein the memory request request includes a memory length of the application; and before the step of determining whether the memory application request meets the permission requirement, the method further includes:

Determining whether the memory length of the application exceeds the total length of the NVM memory, and if not, entering the step of determining whether the memory application request meets the permission requirement; if yes, ending.
The method according to claim 1, wherein the memory request request includes a memory length of the application; and the step of allocating NVM memory according to the application request comprises:

Applying and initializing a virtual address space of the same size according to the memory length of the application, where the virtual address space includes a first address and a tail address of the virtual address space;

Converting the first address of the virtual address space into a physical address, and applying NVM memory of the same size in the NVM according to the physical address and the requested memory length.
The method of claim 3 wherein the virtual address space is located in an indirect mapped address space in a kernel space of Linux.
The method according to claim 1, wherein the physical address of the NVM memory is pre-allocated in a ZONE_HIGHMEN area in a Linux kernel memory management area, and is connected to a ZONE_NORMAL area in a Linux kernel memory management area.
A memory management system based on heterogeneous mixed memory, characterized in that the system comprises:

a receiving module, configured to receive a memory application request;

a detecting module, configured to detect whether an NVM flag exists in the memory application request;

a first determining module, configured to determine, if the NVM flag exists in the memory application request, whether the memory application request meets a permission requirement, and if yes, allocate NVM memory according to the memory application request, and if not, not process ;

An application module, configured to allocate NVM memory according to the memory application request if the memory application request meets a permission requirement;

The application module is further configured to allocate DRAM memory according to the memory request request if the NVM flag does not exist in the memory application request.
The system according to claim 6, wherein the memory request request includes a memory length of the application, and the system further comprises:

The second judging module is configured to determine whether the memory length of the application exceeds the total length of the NVM memory, and if not, notify the first judging module whether the memory application request meets the permission requirement, and if not, not deal with.
The system according to claim 6, wherein the memory application request includes a memory length of the application, and the application module includes:

a virtual address space application unit, configured to apply for the same size virtual number according to the memory length of the application The address space is initialized, and the virtual address space includes a first address and a tail address of the virtual address space;

An address conversion unit, configured to convert a first address of the virtual address space into a physical address;

The memory application unit is configured to apply for the same size of the NVM memory in the NVM according to the physical address and the requested memory length.
The system of claim 8 wherein said virtual address space is located in an indirect mapped address space in a kernel space of Linux.
The system according to claim 6, wherein the physical address of the NVM memory is pre-allocated in the ZONE_HIGHMEN area in the Linux kernel memory management area, and is connected to the ZONE_NORMAL area in the Linux kernel memory management area.