#第七次实验报告 ###1. 简介深入理解操作系统的段、页式内存管理，深入理解段表、页表、逻辑地址、线性地址、物理地址。 ###2. 实验内容 ####2.1. 跟踪地址翻译通过地址ds:offset 查找目标物理地址的步骤列如下： - 通过gdtr寄存器（存放物理地址）+ ldtr（存放选择子），来获得进程ldt的物理地址。 - 再根据段选择器（DS/CS）定位，段选择符在LDT中的位置，用以获得线性地址的基地址，其中LDT和GDT中的选择符结构一致。 - 得到的线性基地址+段内偏移量即为线性地址。 - 根据线性地址利用页表映射成物理地址，首先根据CR3寄存器获得页表项目录（page dir）基地址 . - 根据线性地址的前10bit获取page table 在page dir中的位置。获取page table的物理地址，注意对于该物理的第三位需要ignor。 - 再根据线性地址的12-21bit来获取物理页项在页表（page table）中的位置。获取物理页后，再根据线性地址中的低12位获取，目标数据。寻址完成。 #####2.2.1 实验结果 ![图片描述](https://dn-simplecloud.shiyanlou.com/courses/uid523698-20190714-1563105409474) #####2.2.2 实验问题 - 最重要的四步： 1. 通过段寄存器确定查询的是LDT表或者是GDT表 2. 结合ldtr和ldtr寄存器来获得线性地址 3. 根据CR3中的目录基址获得线性地址对应的物理地址 - 在执行一次test会有的异同： 1. 段基址可能会变化，因为段基址是根据进程次序由操作系统分配的64M空间。 2. 数据段内偏移量不会偏移量[逻辑地址]不会变，因为这是编译后就不变了。 ####2.2 增加共享内存功能，并将生产者—消费者程序移植到 Linux 0.11 #####2.2.1 要点 - 难点在于获得进程可用的线性地址，一个进程64M线性空间内，在低地址端存放了进程代码，数据，和bss段，高地址端存放了环境变量，用户栈; ![图片描述](https://dn-simplecloud.shiyanlou.com/courses/uid523698-20190728-1564279787060) - 根据上图，函数shmat将从bss段末处开始查找一页空闲的线性地址，即current->brk; - 返回的地址应当是逻辑地址，所以不应是完整的线性地址，而是相对于段基址的段内偏移； - 共享内存在put进进程的线性地址空间后，需要将mm_map中的引用递增1。 #####2.2.2 实验代码 - shm.c 增加2个系统调用函数，分别用于分配物理页空间，和绑定用户线性地址空间。 ```c #include #include #include #include #include #define MAX_PAGES 5 #define PAGE_SIZE 4096 typedef struct { key_t key; int shmid; unsigned long page; }shm_t; shm_t shm[5] = { {0,0,0},{0,0,0},{0,0,0},{0,0,0},{0,0,0} }; int _shm_counter = 0; int sys_shmget(key_t key, size_t size, int shmflg) { int i,j; if (!key) return -1; for (i = 0,j= MAX_PAGES; i < MAX_PAGES; i++) { if (shm[i].key == key) break; if (shm[i].key == 0)// available j = i; } if (i < MAX_PAGES) return shm[i].shmid; else if (j == MAX_PAGES) return -1; //no idle key-page space found if (size > PAGE_SIZE) return -EINVAL; shm[j].page = get_free_page(); if (!shm[j].page) return -ENOMEM; shm[j].key = key; shm[j].shmid = ++_shm_counter; return shm[j].shmid; } void* sys_shmat(int shmid, const void* shmaddr, int shmflg) { //to more high level int i; unsigned long* page_table; unsigned long* page_dir; unsigned long offset,table_off; for (i = 0; i < MAX_PAGES; i++) { if (shm[i].shmid == shmid) break; } if(i == MAX_PAGES) return (void *)(-EINVAL); //find idle page table item /* page_dir = (unsigned long*)((current->start_code >> 20) & 0xffc); size = (0x4000000 + 0x3fffff) >> 22; dir_off = 0; table_off = 0; for (; size-- > 0; page_dir++) { if (!(*page_dir & 1)) continue; page_table = (unsigned long*)(0xfffff000 & *page_dir); for (nr = 0; nr < 1024;nr++) { if (!(1 & *page_table)) { put_page(shm[i].page, current->start_code+ dir_off+table_off); return (void*)((dir_off + table_off) & 0xfffff000); } page_table++; table_off += 0x1000; } dir_off += 0x400000; } */ offset = current->brk; for (offset = current->brk; offset < 0x400000; offset = offset + PAGE_SIZE) { page_dir = (unsigned long*)(((current->start_code + offset) >> 20) & 0xffc); if (!(*page_dir & 1)) continue; page_table = (unsigned long*)(0xfffff000 & *page_dir); table_off = ((current->start_code + offset) >> 12) & 0x03FF; if (!(1 & *(page_table + table_off * 4))) { if (put_shm_page(shm[i].page, current->start_code + offset) < 0) return (void *)-1; else return offset; } } return (void*)-1; } ``` - mm/memory 增加put_shm_page 用于将共享内存页增加到线性地址空间内 ```c int put_shm_page(unsigned long page, unsigned long address) { unsigned long tmp, * page_table; /* NOTE !!! This uses the fact that _pg_dir=0 */ if (page < LOW_MEM || page >= HIGH_MEMORY) printk("Trying to put page %p at %p\n", page, address); page_table = (unsigned long*)((address >> 20) & 0xffc); if ((*page_table) & 1) page_table = (unsigned long*)(0xfffff000 & *page_table); else { if (!(tmp = get_free_page())) return -1; *page_table = tmp | 7; page_table = (unsigned long*)tmp; } page_table[(address >> 12) & 0x3ff] = page | 7; /* no need for invalidate */ page -= LOW_MEM; (mem_map[page >>12])++; return 0; } ``` - sem.c 也做了部分改动，因为跟上次实验不同的是，消费者和生产者在不同的文件中执行，所以有俩次unlink 信号量，所以加了计数器，到0才释放 ```c /* * linux/kernel/sem.c * (C) 2019 Tsai Tsu Quen */ #include #include #include #include #include sem_t sem_list[NR_SEM]; wait_node_t wait_node_list[NR_WAIT_NODE]; wait_node_t* _find_idle_wait_node() { int i = 0; for (i = 0; i < NR_WAIT_NODE; i++) { if (wait_node_list[i].idle == 1) { wait_node_list[i].idle = 0; wait_node_list[i].next_node = NULL; return (wait_node_t*)(wait_node_list + i); } } printk("no wait node\n"); return NULL; } int _str_compare(char* buf_1, char* buf_2) { int i = 0; while ((buf_1[i] == buf_2[i]) && (buf_1[i] != '\0')) { i++; } if (buf_1[i] == '\0' && buf_2[i] == '\0') return 0; else return -1; } int _str_copy(char* buf_s, char* buf_d) { int i = 0; do { buf_d[i] = buf_s[i]; i++; } while (buf_s[i] != '\0'); return 0; } sem_t* sys_sem_open(const char* name, int value) { int i = 0; int j = NR_SEM + 1; char name_buf[20]; while ((name_buf[i] = get_fs_byte(name + i)) != '\0') i++; for (i = 0; i < NR_SEM; i++) { if (*(sem_list[i].name) == '\0') j = i; else if (_str_compare(sem_list[i].name, name_buf) == 0) { sem_list[i].counter++; return (sem_t*)(sem_list + i); } } if (j < NR_SEM) { _str_copy(name_buf, sem_list[j].name); sem_list[j].counter = 1; sem_list[j].value = value; sem_list[j].p_wait_queue = NULL; return (sem_t*)(sem_list + j); } else return NULL; } int sys_sem_unlink(const char* name) { int i = 0; char name_buf[20]; while ((name_buf[i] = get_fs_byte(name + i)) != '\0') i++; for (i = 0; i < NR_SEM; i++) { if (_str_compare(sem_list[i].name, name_buf) == 0) { sem_list[i].counter--; if (sem_list[i].counter == 0) { (sem_list[i].name)[0] = '\0'; sem_list[i].p_wait_queue = NULL; } return 0; } } return -1; } void sem_init() { int i; for (i = 0; i < NR_SEM; i++) { (sem_list[i].name)[0] = '\0'; sem_list[i].counter = 0; } for (i = 0; i < NR_WAIT_NODE; i++) { (wait_node_list[i]).idle = 1; } return; } int sys_sem_wait(sem_t * sem) { wait_node_t* p, * q; if (sem == NULL) return -1; cli(); sem->value--; //printk("In wait ,sem %s value is %d\n",sem->name,sem->value); if (sem->value < 0) { current->state = TASK_UNINTERRUPTIBLE; p = _find_idle_wait_node(); p->p_task_struct = current; if (sem->p_wait_queue) { q = sem->p_wait_queue; while (q->next_node) q = q->next_node; q->next_node = p; } else sem->p_wait_queue = p; schedule(); } sti(); return 0; } int sys_sem_post(sem_t * sem) { wait_node_t* p; if (sem == NULL) return -1; cli(); sem->value++; //printk("In post sem %s value is %d\n",sem->name,sem->value); if (sem->value <= 0) { /*p = sem->sem_queue[i];*/ p = sem->p_wait_queue; sem->p_wait_queue = p->next_node; (p->p_task_struct)->state = 0; p->idle = 1; p->next_node = NULL; } sti(); return 0; } ``` - producer 共享内存页作为载体。 ```c #define __LIBRARY__ #include #include #include #include #define BUFFER_SIZE 10 #define NUMS 500 _syscall2(sem_t*, sem_open, const char*, name, int, value); _syscall1(int, sem_wait, sem_t*, sem); _syscall1(int, sem_post, sem_t*, sem); _syscall1(int, sem_unlink, const char*, name); _syscall3(void *, shmat,int ,shmid,const void *,shmaddr,int,shmflg); _syscall3(int, shmget, key_t, key, size_t, size, int, shmflg); int main(void) { int num,counter = 0; int state; int off = 0; int shmid; int* p_shm; key_t shmkey = 1225; sem_t* Empty, * Full, * Mutex; if (!fork()) { if ((Empty = sem_open("Empty", BUFFER_SIZE)) == NULL) return -1; if ((Full = sem_open("Full", 0)) == NULL) return -1; if ((Mutex = sem_open("Mutex", 1)) == NULL) return -1; shmid = shmget(shmkey, 4096, 0); if (shmid < 0) { printf("error : %d", shmid); return 0; } p_shm = (int*)shmat(shmid, (const void*)0, 0); if ((int)p_shm < 0) { printf("error: %d,", ((int)p_shm)); return 0; } while (counter < NUMS) { sem_wait(Empty); sem_wait(Mutex); num = counter; p_shm[counter % BUFFER_SIZE] = num; counter++; sem_post(Mutex); sem_post(Full); } if (sem_unlink("Empty") < 0) return -1; if (sem_unlink("Full") < 0) return -1; if (sem_unlink("Mutex") < 0) return -1; return 0; } while (wait(&state) > 0); return 0; } ``` - consumer 使用共享内存 ```c #define __LIBRARY__ #include #include #include #include #define BUFFER_SIZE 10 #define NUMS 500 _syscall2(sem_t*, sem_open, const char*, name, int, value); _syscall1(int, sem_wait, sem_t*, sem); _syscall1(int, sem_post, sem_t*, sem); _syscall1(int, sem_unlink, const char*, name); _syscall3(void*, shmat, int, shmid, const void*, shmaddr, int, shmflg); _syscall3(int, shmget, key_t, key, size_t, size, int, shmflg); int main(void) { int i; int state; int shmid; int* p_shm; key_t shmkey = 1225; FILE * fp; sem_t* Empty, * Full, * Mutex; if (!fork()) { fp = freopen("output.txt", "w", stdout); if ((Empty = sem_open("Empty", BUFFER_SIZE)) == NULL) return -1; if ((Full = sem_open("Full", 0)) == NULL) return -1; if ((Mutex = sem_open("Mutex", 1)) == NULL) return -1; shmid = shmget(shmkey, 4096, 0); if (shmid < 0) { printf("error : %d", shmid); return 0; } p_shm = (int*)shmat(shmid, (const void*)0, 0); if ((int)p_shm < 0) { printf("error: %d,", ((int)p_shm)); return 0; } for (i = 0; i < NUMS; i++) { sem_wait(Full); sem_wait(Mutex); printf("%d : %d\n", getpid(), p_shm[i % BUFFER_SIZE]); sem_post(Mutex); sem_post(Empty); } fclose(fp); if (sem_unlink("Empty") < 0) return -1; if (sem_unlink("Full") < 0) return -1; if (sem_unlink("Mutex") < 0) return -1; return 0; } while (wait(&state) > 0); return 0; } ``` #####2.2.4 实验结果 ![图片描述](https://dn-simplecloud.shiyanlou.com/courses/uid523698-20190728-1564281044490) #####2.2.5 遗留问题共享页的物理内存并不会在进程结束后释放，事实上，在进程的exit中中执行free_pages(LDT(data))，但只会将mem_map对应物理地址的计数器减一，。可能造成memory leaks.

“操作系统原理与实践”实验报告