CN100383739C - Startup Optimization Method for Embedded Operating System Image Startup - Google Patents

Abstract

The invention discloses a startup optimization method for mirror startup of an embedded operating system. The method that the present invention adopts is that when the operating system is started, the boot program directly reads the original saved system minimum task image into the memory from the memory, and sets the attributes of various structures and the information of the system register memory according to the information read in by the system. Quickly restore the system to the working state to achieve a quick start method. The invention proposes a start-up optimization method for operating system image start-up, which improves the original start-up process of the operating system, making the system start-up faster and reaching the working state of the system very quickly.

Description

Startup Optimization Method for Embedded Operating System Image Startup

technical field

The invention relates to the field of startup based on an embedded operating system, in particular to a startup optimization method for mirror startup of an embedded operating system.

Background technique

Embedded systems such as mobile phones cannot take a long time to start because of human tolerance. For the convenience of use, the boot and startup time of the mobile phone operating system should be shortened as much as possible. Although the kernel has been greatly reduced and various functional modules have been optimized as much as possible, due to hardware constraints, if the boot and startup For special design, it will still cause the startup time to be too long.

Now mainly adopt two kinds of techniques to shorten this time. First, solidify the settings of various non-extensible hardware, try to standardize the interface of extensible equipment, and save equipment detection time. 2. Delayed loading technology, LazyLoading, this technology does not load modules that are not currently required. These modules can be loaded during the standby time after startup, or after they are specifically used, to break up the startup time, thereby shortening the time spent on the module. Reaction time is critical for users.

For the first technology, at present and in a short period of time in the future, the most important thing is to solidify the settings of non-extensible hardware, because first of all, there are few scalable devices for mobile phones, and secondly, there are many hardware conditions, so some It should be an ideal measure to fix the fixed settings inside the operating system. And considering the future development trend, the mobile phone will be able to connect more and more peripherals, and these peripherals must be plug and play, so it is impossible to fix their settings. However, if a set of standard interfaces is established for all mobile phone peripherals, at most one standard public interface module only needs to be loaded during startup, which greatly saves startup time and storage space.

For the second technology, since the operating system will adopt "componentized" technology in many places, it is very convenient for Lazy Loading. And in order to further shorten the response time, "dynamically set" the necessary modules, according to the statistics of users' usage habits and other conditions, set some modules that users will not use immediately after starting up as non-essential modules, and set users who often The functions that will be used later are set as mandatory modules, so that the reaction time for the user is greatly improved.

Contents of the invention

The purpose of the present invention is to provide a startup optimization method for embedded operating system image startup.

The technical scheme that the present invention solves its technical problem adopts is as follows:

1) Operating system image preparation:

During the startup process, the boot program of the operating system first copies itself from the external storage to the memory space, and then searches for the operating system kernel image at the specified external storage location. If the operating system kernel image exists, then copy the operating system kernel image from the The specified external memory location is loaded into the kernel and decompressed. During the startup process, the operating system needs to judge whether there is a fast-boot operating system image. If it does not exist, it will start normally. If it exists, it will start quickly. There is an original operating system image, and the operating system will start according to the normal procedure;

When the operating system starts to the simplest basic working state, the operating system will start the operating system hibernation process. The main job of this process is to make various preparations for saving the operating system image;

2) Save the operating system image:

This part mainly completes the on-site preservation of the operating system. First, a virtual terminal will be established to simulate the communication between the kernel and the user mode process, and then start to end all processes except the kernel mode process, the dead process, and the current process, and then call Function to release as much memory space as possible to reduce the size of the operating system image, then process some drivers, send a suspend command to all peripherals registered in the power management system, then insert a memory barrier, and start saving processing server context, and finally mirror the operating system to an external specified device, and finally sleep the operating system;

3) Loading of operating system image:

During the startup process, the boot program of the operating system first copies itself from the external storage to the memory space, and then searches for the operating system kernel image at the specified external storage location. If the operating system kernel image exists, then copy the operating system kernel image from the The specified external storage location is loaded into the kernel and decompressed. During the startup process, if the boot program finds that there is a fast-boot operating system image, the operating system loads the image into the memory and decompresses it;

The operating system first creates a virtual terminal to display system information during the process of loading the image, first checks the correctness of the operating system image, and if it is correct, loads the operating system image, loops to obtain all saved pages, and restores the memory content , restore the value of the processor register through the inserted memory barrier, and finally release the memory space requested during the restoration process, and complete the loading of the entire operating system image by sending a system restoration notification to the device registered in the power management module.

Compared with the background technology, the present invention has the beneficial effects that:

1. No delay: The above method can only create a false appearance for the user. The system seems to be usable and has already started up. In fact, the necessary modules must be loaded when it is used. The startup optimization method of embedded operating system image startup is to provide a complete system for the application, directly by setting the application environment of the system, without delay loading, and loading the necessary modules when in use, saving time and energy;

2. Quickness: The startup optimization method of embedded operating system image startup adopts the startup method of directly setting the system environment, which is faster;

3. Versatility: The startup image can be set automatically, and there is no need to make different settings for different application platforms. It is not necessary to change the image when changing the platform, just restart the system, and then let the system automatically set an image according to the system situation.

Description of drawings

Accompanying drawing is the flowchart of whole system.

Detailed ways

When implementing boot optimization methods based on embedded operating systems, mirror boot technology is widely used.

The specific implementation process of the startup optimization method for mirror startup of the embedded operating system is as follows.

1. System image preparation

The system image preparation process is mainly divided into the following parts:

1. Operating system boot

During the boot process of the operating system, the boot program first copies itself from the external storage to the internal memory space, and then searches for the system image at the specified external storage location. If the system image exists, the operating system kernel image is copied from the external storage of the system. Upload and load the kernel. Since the system kernel image is relatively large, it is usually compressed and saved to the external memory, so it needs to be decompressed;

Taking the ARM-Linux operating system on the Sistang embedded experimental platform as an example, its operating system boot process can be seen below.

Before the kernel runs, the system bootloader needs to complete the loading of the kernel and some auxiliary work, and then jump to the starting address of the kernel code and execute it.

First of all, after booting, the Sitsang board maps the address 0x0c00 0000 to 0 (can be set by jumpers), actually boots from 0x0c00 0000, and enters the bootloader, but because the flash speed is slow, there is a small program in front of the bootloader to copy the bootloader to SDRAM 0x0AFE0100, and then run the bootloader from 0x 0800 0000, call this small program flashloader, flashloader must first initialize SDRAM. The flashloader loads the bootloader to 0x0AFE0100, and then jumps back. In fact, 0x0AFE0100 is the real bootloader burned in the flash 0x0C000100.

BootLoader transfers the operating system code into the memory, and then hands over control to the Setup.S program in arch/arm/boot. This program of Setup.S performs basic detection and settings on the system in real mode, then transfers to protection mode and hands over the control right to head_armv.S. One step is the decompression process of the system kernel. This part of the code is at address 0x1000 (file/Boot/head.S). This program initializes the registers and then executes decompress_kernel(). This function originates from zBoot/inflate.c, zBoot/unzip .c and zBoot/misc.c three files. After decompression, the data is placed at address 0x100000. Head.S establishes the framework of memory management and interrupt management.

2. Start the operating system

1), determine whether there is a quick boot image

After the kernel image is decompressed, the system needs to judge whether there is a fast boot image, if it does not exist, it will start normally, if it exists, it will start quickly.

2), normal startup

During the boot process, if the boot program finds that there is only the original operating system image, the system will start according to the normal program. First, the system will initialize various components in order, first initialize the architecture, and then initialize the interrupt vector table, scheduler, memory, etc. Wait, and finally start various services of the system.

Taking ARM-Linux as an example, the startup process mainly starts from Start_kernel until the init process is created. After the Init process calls lock_kernel, before calling prepare_namespace to load rootfs, call the do_basic_setup function to perform some basic settings, such as pci, sbus, ecard, etc., do_basic_setup then calls start_context_thread() and do_initcalls() to load some device drivers, and then , enter the newly added function software_resume(), and prepare to check the swap partition and restore the system.

After the software_resume function checks some necessary conditions, it calls the read_suspend_image function to read the previously saved image. If the image does not exist and fails to read, the software_resume function calls kernel_thread(swsusp_mainloop, NULL, CLONE_FS|CLONE_FILES|CLONE_SIGHAND|SIGCHLD) to create A kernel mode process, the swsusp_mainloop() function calls daemonize() to release the reference to the user space to make itself a background daemon process, and rename itself to kswsuspd, and then falls into an infinite loop, and detects every second Check whether the value of swsusp_state[0] changes (if the value changes from 0 to 1, it will jump out of the loop and call the sleep function). At this point, the software_resume function returns, continue do_basic_setup and init(), until the system starts normally.

3), mirror start

If the boot program finds that there is an image of the operating system, the boot program loads the image into memory and decompresses it; the system first creates a virtual terminal to display system information during the process of loading the image, first checks the correctness of the image, and then The memory pages in the image are restored to the memory in order, and finally the context of the processor is restored. The specific examples of this part of the work will be put later.

3. Establishment of a mirror working environment

When the system starts to the simplest basic working state, the operating system will start the system hibernation process. The main job of this process is to make various preparations for saving the system image;

Taking the image startup of the Linux operating system as an example, this part of the work mainly includes:

1) Add a section to the data segment of the kernel: _nosave, so as to put some variables in it. In this way, when using these variables, the stack will not be involved.

2) Define the device static char resume_file[256]="/dev/modify" for saving/restoring the mirror image, the major device number of this device is 241, the minor device number is 2, and root_dev_names[ in init/do_mounts.c is required at the same time ], add an item: {"modify", 0xf102}, after that, through the name_to_kdev_t function, specify the parameter "/dev/modify" to get the device number 0xf102 of the device, and then use the macro MAJOR() and MINOR() to respectively Get the device's major device number 241 and slave device number 2.

3) Define the state quantity: int swsusp_state[3] is used to manage some parameter information in the process of dormancy and mirroring.

2. Save the system image

This part mainly completes the preservation work of the system site. First, a virtual terminal will be established to simulate communication; memory space to reduce the size of the memory image, and then process some drivers, send a suspend command to all peripherals registered in the power management system, and then insert a system barrier, block system execution, and start saving the processor context , and finally mirror the memory to an externally specified device, and finally sleep the operating system.

Taking ARM-Linux as an example, the whole process is as follows:

1), the system must call a function called software_suspend to create a mirror image. software_suspend() calls the do_software_suspend function, which really starts the system's hibernation and image creation.

2), software_suspend() calls the do_software_suspend function, first creates a virtual terminal, and then starts calling the freeze_processes function to end all processes except the kernel mode process, the zombie process, and the current process, and then calls the free_some_memory function to release as much as possible memory space to reduce the size of the memory image. Then, some drivers are processed, and pm_send_all(PM_SUSPEND, (void*)3) is called to send a system suspend notification to the device registered in the power management PM module.

3). Finally, the do_software_suspend function calls the do_suspend_lowlevel function. do_suspend_lowlevel calls do_magic_suspend_1(), save_processor_context() and do_magic_suspend_2() three functions in turn. do_magic_suspend_1() calls mb() and barrier() to prepare for the next save_processor_context function. Void Barrier (void) informs the compiler to insert a memory barrier, but it is invalid for the hardware. The compiled code will store all the modified values in the current CPU registers into the memory, and then read them out from the memory again when the data is needed. .

4). The save_processor_context function saves the running context of the CPU in the current state. A struct used to hold these values has the attribute _attribute_((packed)), which can compactly arrange its member variables.

5), then enter the do_magic_suspend_2 function. This function is responsible for saving the memory page to the swap partition. This function first calls read_swapfiles to find the partition specified in the previous article from the possible 32 swap devices (including files and partitions): /dev/modify. Then call save_suspend_image to complete all copy operations, call suspend_power_down, shut down or restart the computer.

Third, the loading of the operating system image

During the startup process, the boot program of the operating system loads the kernel image of the operating system from the external memory of the system into the kernel and decompresses it. The image is loaded into memory and decompressed;

The system first creates a virtual terminal to display system information during the process of loading the image, first checks the correctness of the image, then restores the memory pages in the image to the memory in order, and finally restores the context of the processor;

Taking ARM-Linux as an example, the whole process is as follows:

1) After booting, BootLoader loads the linux kernel and decompresses it, then start_kernel until the init process is created. After the Init process calls lock_kernel, before calling prepare_namespace to load rootfs, call the do_basic_setup function to perform some basic settings, such as pci, sbus, ecard, etc., do_basic_setup then calls start_context_thread() and do_initcalls() to load some device drivers, and then , enter your own software_resume() function, ready to check the swap partition and restore the system.

2) After the software_resume function checks some necessary conditions, it calls the read_suspend_image function to read the previously saved image, and read_suspend_image calls the lower-level _read_suspend_image to complete the active image reading work.

3), _read_suspend_image calls prepare_suspend_console to create a virtual console to display some relevant information during recovery. Then call bdev_read_page to get a page from the /dev/modify device, get the header information of the image, and call the sanity_check function to check whether the header information of the image is correct, to determine whether the current system is the system before hibernation, if not Then start according to the normal system.

4) Next, pagedir and all saved memory pages are obtained by calling bdev_read_page cyclically (the order of reading memory pages is just opposite to the order of saving memory pages). And restore the memory content, and then the read_suspend_image function returns. Next call do_suspend_lowlevel to further restore the system.

5), do_suspend_lowlevel changes the physical address of swapper_pg_dir to the physical address of the page directory by setting the CPU register Cr3. And by setting ecx, a new stack pointer is created. CR3 is used to save the starting physical address of the page directory table. Since the directory is page-aligned, only the upper 20 bits are valid, and the lower 12 bits are reserved for unused. When loading a new value into CR3, the lower 12 bits must be 0; but when fetching a value from CR3, the lower 12 bits are ignored. Whenever the value of CR3 is reset with the MOV instruction, the content of the paging mechanism high-speed buffer will be invalid. In this way, the high-speed cache of the paging mechanism can be pre-refreshed before the paging mechanism is enabled, that is, before the PG bit is set to 1. The CR3 register can be loaded even when the PG bit or the PE bit of the CR0 register is 0. For example, CR3 can also be set in the real mode to initialize the paging mechanism. When the task is switched, CR3 will be changed, but if the value of CR3 in the new task is the same as the value of CR3 in the original task, then the processor will not refresh the paging cache, so as to have a faster execution speed when the task shares the same table.

6), do_suspend_lowlevel then calls do_magic_resume_1() restore_processor_context and do_magic_resume_2() in turn to complete all recovery work.

7), do_magic_resume_1() calls mb(); barrier(); to prepare for the next restore_processor_context function. Void Barrier (void) informs the compiler to insert a memory barrier, but it is invalid for the hardware. The compiled code will store all the modified values in the current CPU registers into the memory, and then read them out from the memory again when the data is needed. . After that, do_suspend_lowlevel starts copying memory pages in units of sizeof(char). The reason why functions such as copy_page cannot be used here is that when the memory page is copied, the stack (stack) will be destroyed. Before the value of the CPU register is completely restored, the stack and local variables can no longer be used, but it can be used in the kernel code segment Variables in the _nosave area. After that, the restore_processor_context() function starts to restore the values of the CPU registers. Finally enter do_magic_resume_2().

8), do_magic_resume_2 () mainly deals with some work of cleaning the scene. It releases the pagedir allocated during hibernation to ensure that the memory is completely restored to the state before hibernation, and then sends a system resume notification to the device registered in the power management PM module by calling pm_send_all(PM_RESUME, (void*)0) . At this point, all the work is done. The system has returned to the state it was in before hibernation.

The flowchart of the whole system is shown in the accompanying drawing.

Claims (1)

1. a startup optimization method for embedded operating system image startup, characterized in that: 1) Operating system image preparation: During the startup process, the boot program of the operating system first copies itself from the external storage to the memory space, and then searches for the operating system kernel image at the specified external storage location. If the operating system kernel image exists, then copy the operating system kernel image from the The specified external memory location is loaded into the kernel and decompressed. During the startup process, the operating system needs to judge whether there is a fast-boot operating system image. If it does not exist, it will start normally. If it exists, it will start quickly. There is an original operating system image, and the operating system will start according to the normal procedure; When the operating system starts to the simplest basic working state, the operating system will start the operating system hibernation process. The main job of this process is to make various preparations for saving the operating system image; 2) Save the operating system image: This part mainly completes the on-site preservation of the operating system. First, a virtual terminal will be established to simulate the communication between the kernel and the user mode process, and then start to end all processes except the kernel mode process, the dead process, and the current process, and then call Function to release as much memory space as possible to reduce the size of the operating system image, then process some drivers, send a suspend command to all peripherals registered in the power management system, then insert a memory barrier, and start saving processing server context, and finally mirror the operating system to an external specified device, and finally sleep the operating system; 3) Loading of operating system image: During the startup process, the boot program of the operating system first copies itself from the external storage to the memory space, and then searches for the operating system kernel image at the specified external storage location. If the operating system kernel image exists, then copy the operating system kernel image from the The specified external storage location is loaded into the kernel and decompressed. During the startup process, if the boot program finds that there is a fast-boot operating system image, the operating system loads the image into the memory and decompresses it; The operating system first creates a virtual terminal to display system information during the process of loading the image, first checks the correctness of the operating system image, and if it is correct, loads the operating system image, loops to obtain all saved pages, and restores the memory content , restore the value of the processor register through the inserted memory barrier, and finally release the memory space requested during the restoration process, and complete the loading of the entire operating system image by sending a system restoration notification to the device registered in the power management module.

Images