#include #include #include #include #include #include #include #include #include #include #include #include #include #include #define MAX_PROCESS 512 volatile struct proc_info* __current; struct proc_info dummy; struct scheduler sched_ctx; LOG_MODULE("SCHED") void sched_init() { size_t pg_size = ROUNDUP(sizeof(struct proc_info) * MAX_PROCESS, 0x1000); for (size_t i = 0; i <= pg_size; i += 4096) { uintptr_t pa = pmm_alloc_page(KERNEL_PID, PP_FGPERSIST); vmm_set_mapping( PD_REFERENCED, PROC_START + i, pa, PG_PREM_RW, VMAP_NULL); } sched_ctx = (struct scheduler){ ._procs = (struct proc_info*)PROC_START, .ptable_len = 0, .procs_index = 0 }; } void run(struct proc_info* proc) { if (!(__current->state & ~PROC_RUNNING)) { __current->state = PROC_STOPPED; } proc->state = PROC_RUNNING; /* 将tss.esp0设置为上次调度前的esp值。 当处理信号时,上下文信息是不会恢复的,而是保存在用户栈中,然后直接跳转进位于用户空间的sig_wrapper进行 信号的处理。当用户自定义的信号处理函数返回时,sigreturn的系统调用才开始进行上下文的恢复(或者说是进行 另一次调度。 由于这中间没有进行地址空间的交换,所以第二次跳转使用的是同一个内核栈,而之前默认tss.esp0的值是永远指向最顶部 这样一来就有可能会覆盖更早的上下文信息(比如嵌套的信号捕获函数) */ tss_update_esp(proc->intr_ctx.registers.esp); apic_done_servicing(); asm volatile("pushl %0\n" "jmp switch_to\n" ::"r"(proc)); // kernel/asm/x86/interrupt.S } void schedule() { if (!sched_ctx.ptable_len) { return; } // 上下文切换相当的敏感!我们不希望任何的中断打乱栈的顺序…… cpu_disable_interrupt(); struct proc_info* next; int prev_ptr = sched_ctx.procs_index; int ptr = prev_ptr; // round-robin scheduler do { ptr = (ptr + 1) % sched_ctx.ptable_len; next = &sched_ctx._procs[ptr]; } while (next->state != PROC_STOPPED && ptr != prev_ptr); sched_ctx.procs_index = ptr; run(next); } static void proc_timer_callback(struct proc_info* proc) { proc->timer = NULL; proc->state = PROC_STOPPED; } __DEFINE_LXSYSCALL1(unsigned int, sleep, unsigned int, seconds) { // FIXME: sleep的实现或许需要改一下。专门绑一个计时器好像没有必要…… if (!seconds) { return 0; } if (__current->timer) { return __current->timer->counter / timer_context()->running_frequency; } struct lx_timer* timer = timer_run_second(seconds, proc_timer_callback, __current, 0); __current->timer = timer; __current->intr_ctx.registers.eax = seconds; __current->state = PROC_BLOCKED; schedule(); } __DEFINE_LXSYSCALL1(void, exit, int, status) { terminate_proc(status); } __DEFINE_LXSYSCALL(void, yield) { schedule(); } pid_t _wait(pid_t wpid, int* status, int options); __DEFINE_LXSYSCALL1(pid_t, wait, int*, status) { return _wait(-1, status, 0); } __DEFINE_LXSYSCALL3(pid_t, waitpid, pid_t, pid, int*, status, int, options) { return _wait(pid, status, options); } pid_t _wait(pid_t wpid, int* status, int options) { pid_t cur = __current->pid; int status_flags = 0; struct proc_info *proc, *n; if (llist_empty(&__current->children)) { return -1; } wpid = wpid ? wpid : -__current->pgid; cpu_enable_interrupt(); repeat: llist_for_each(proc, n, &__current->children, siblings) { if (!~wpid || proc->pid == wpid || proc->pgid == -wpid) { if (proc->state == PROC_TERMNAT && !options) { status_flags |= PROCTERM; goto done; } if (proc->state == PROC_STOPPED && (options & WUNTRACED)) { status_flags |= PROCSTOP; goto done; } } } if ((options & WNOHANG)) { return 0; } // 放弃当前的运行机会 sched_yield(); goto repeat; done: cpu_disable_interrupt(); *status = (proc->exit_code & 0xffff) | status_flags; return destroy_process(proc->pid); } struct proc_info* alloc_process() { pid_t i = 0; for (; i < sched_ctx.ptable_len && sched_ctx._procs[i].state != PROC_DESTROY; i++) ; if (i == MAX_PROCESS) { panick("Panic in Ponyville shimmer!"); } if (i == sched_ctx.ptable_len) { sched_ctx.ptable_len++; } struct proc_info* proc = &sched_ctx._procs[i]; memset(proc, 0, sizeof(*proc)); proc->state = PROC_CREATED; proc->pid = i; proc->created = clock_systime(); proc->pgid = proc->pid; llist_init_head(&proc->mm.regions); llist_init_head(&proc->children); llist_init_head(&proc->grp_member); return proc; } void commit_process(struct proc_info* process) { assert(process == &sched_ctx._procs[process->pid]); if (process->state != PROC_CREATED) { __current->k_status = LXINVL; return; } // every process is the child of first process (pid=1) if (process->parent) { llist_append(&process->parent->children, &process->siblings); } else { process->parent = &sched_ctx._procs[0]; } process->state = PROC_STOPPED; } // from extern void __del_pagetable(pid_t pid, uintptr_t mount_point); pid_t destroy_process(pid_t pid) { int index = pid; if (index <= 0 || index > sched_ctx.ptable_len) { __current->k_status = LXINVLDPID; return; } struct proc_info* proc = &sched_ctx._procs[index]; proc->state = PROC_DESTROY; llist_delete(&proc->siblings); struct mm_region *pos, *n; llist_for_each(pos, n, &proc->mm.regions.head, head) { lxfree(pos); } vmm_mount_pd(PD_MOUNT_1, proc->page_table); __del_pagetable(pid, PD_MOUNT_1); vmm_unmount_pd(PD_MOUNT_1); return pid; } void terminate_proc(int exit_code) { __current->state = PROC_TERMNAT; __current->exit_code = exit_code; __SET_SIGNAL(__current->parent->sig_pending, _SIGCHLD); schedule(); } struct proc_info* get_process(pid_t pid) { int index = pid; if (index < 0 || index > sched_ctx.ptable_len) { return NULL; } return &sched_ctx._procs[index]; } int orphaned_proc(pid_t pid) { if (!pid) return 0; if (pid >= sched_ctx.ptable_len) return 0; struct proc_info* proc = &sched_ctx._procs[pid]; struct proc_info* parent = proc->parent; // 如果其父进程的状态是terminated 或 destroy中的一种 // 或者其父进程是在该进程之后创建的,那么该进程为孤儿进程 return (parent->state & PROC_TERMMASK) || parent->created > proc->created; }