refactor: clean up the virtual memory mappings

[lunaix-os.git] / lunaix-os / kernel / sched.c

diff --git a/lunaix-os/kernel/sched.c b/lunaix-os/kernel/sched.c
index 758b157befb68a2b48656ab90f9887067c0f4997..9061e8032c156bcdd61d7c5b30df2b73e66e4dba 100644 (file)
--- a/lunaix-os/kernel/sched.c
+++ b/lunaix-os/kernel/sched.c
@@ -1,226 +1,393 @@
-#include <lunaix/process.h>
-#include <lunaix/sched.h>
-#include <lunaix/mm/vmm.h>
-#include <lunaix/mm/kalloc.h>
-#include <hal/cpu.h>

 #include <arch/x86/interrupts.h>
 #include <arch/x86/tss.h>
 #include <arch/x86/interrupts.h>
 #include <arch/x86/tss.h>

+

 #include <hal/apic.h>
 #include <hal/apic.h>

+#include <hal/cpu.h>

 
 

+#include <lunaix/mm/cake.h>
+#include <lunaix/mm/kalloc.h>
+#include <lunaix/mm/pmm.h>
+#include <lunaix/mm/valloc.h>
+#include <lunaix/mm/vmm.h>
+#include <lunaix/process.h>
+#include <lunaix/sched.h>
+#include <lunaix/signal.h>

 #include <lunaix/spike.h>
 #include <lunaix/status.h>
 #include <lunaix/spike.h>
 #include <lunaix/status.h>

-#include <lunaix/syslog.h>

 #include <lunaix/syscall.h>
 #include <lunaix/syscall.h>

+#include <lunaix/syslog.h>

 
 

-#define MAX_PROCESS 512
+#define PROC_TABLE_SIZE 8192
+#define MAX_PROCESS (PROC_TABLE_SIZE / sizeof(uintptr_t))

 
 volatile struct proc_info* __current;
 
 struct proc_info dummy;
 
 
 volatile struct proc_info* __current;
 
 struct proc_info dummy;
 

-extern void __proc_table;
-

 struct scheduler sched_ctx;
 
 struct scheduler sched_ctx;
 

+struct cake_pile* proc_pile;
+

 LOG_MODULE("SCHED")
 
 LOG_MODULE("SCHED")
 

-void sched_init() {
-    size_t pg_size = ROUNDUP(sizeof(struct proc_info) * MAX_PROCESS, 0x1000);
-    assert_msg(
-        vmm_alloc_pages(KERNEL_PID, &__proc_table, pg_size, PG_PREM_RW, PP_FGPERSIST), 
-        "Fail to allocate proc table"
-    );
-    
-    sched_ctx = (struct scheduler) {
-        ._procs = (struct proc_info*) &__proc_table,
-        .ptable_len = 0,
-        .procs_index = 0
-    };
+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);
+    // }
+
+    proc_pile = cake_new_pile("proc", sizeof(struct proc_info), 1, 0);
+    cake_set_constructor(proc_pile, cake_ctor_zeroing);
+
+    sched_ctx = (struct scheduler){ ._procs = vzalloc(PROC_TABLE_SIZE),
+                                    .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;
-    
-    // FIXME: 这里还是得再考虑一下。
-    // tss_update_esp(__current->intr_ctx.esp);
+void
+run(struct proc_info* proc)
+{
+    proc->state = PS_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);

 
 

-    if (__current->page_table != proc->page_table) {
-        __current = proc;
-        cpu_lcr3(__current->page_table);
-        // from now on, the we are in the kstack of another process
-    }
-    else {
-        __current = proc;
+    apic_done_servicing();
+
+    asm volatile("pushl %0\n"
+                 "jmp switch_to\n" ::"r"(proc)
+                 : "memory"); // kernel/asm/x86/interrupt.S
+}
+
+int
+can_schedule(struct proc_info* proc)
+{
+    if (__SIGTEST(proc->sig_pending, _SIGCONT)) {
+        __SIGCLEAR(proc->sig_pending, _SIGSTOP);
+    } else if (__SIGTEST(proc->sig_pending, _SIGSTOP)) {
+        // 如果进程受到SIGSTOP，则该进程不给予调度。
+        return 0;

     }
 
     }
 

-    apic_done_servicing();
+    return 1;
+}
+
+void
+check_sleepers()
+{
+    struct proc_info* leader = sched_ctx._procs[0];
+    struct proc_info *pos, *n;
+    time_t now = clock_systime();
+    llist_for_each(pos, n, &leader->sleep.sleepers, sleep.sleepers)
+    {
+        if (PROC_TERMINATED(pos->state)) {
+            goto del;
+        }
+
+        time_t wtime = pos->sleep.wakeup_time;
+        time_t atime = pos->sleep.alarm_time;
+
+        if (wtime && now >= wtime) {
+            pos->sleep.wakeup_time = 0;
+            pos->state = PS_READY;
+        }
+
+        if (atime && now >= atime) {
+            pos->sleep.alarm_time = 0;
+            __SIGSET(pos->sig_pending, _SIGALRM);
+        }

 
 

-    asm volatile (
-        "pushl %0\n"
-        "jmp soft_iret\n"::"r"(&__current->intr_ctx): "memory");
+        if (!wtime && !atime) {
+        del:
+            llist_delete(&pos->sleep.sleepers);
+        }
+    }

 }
 
 }
 

-void schedule() {
+void
+schedule()
+{

     if (!sched_ctx.ptable_len) {
         return;
     }
 
     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;
     struct proc_info* next;
     int prev_ptr = sched_ctx.procs_index;
     int ptr = prev_ptr;

+
+    if (!(__current->state & ~PS_RUNNING)) {
+        __current->state = PS_READY;
+    }
+
+    check_sleepers();
+

     // round-robin scheduler
     // round-robin scheduler

+redo:

     do {
         ptr = (ptr + 1) % sched_ctx.ptable_len;
     do {
         ptr = (ptr + 1) % sched_ctx.ptable_len;

-        next = &sched_ctx._procs[ptr];
-    } while(next->state != PROC_STOPPED && ptr != prev_ptr);
-    
+        next = sched_ctx._procs[ptr];
+    } while (!next || (next->state != PS_READY && ptr != prev_ptr));
+

     sched_ctx.procs_index = ptr;
 
     sched_ctx.procs_index = ptr;
 

+    if (!can_schedule(next)) {
+        // 如果该进程不给予调度，则尝试重新选择
+        goto redo;
+    }

 
     run(next);
 }
 
 
     run(next);
 }
 

-static void proc_timer_callback(struct proc_info* proc) {
-    proc->timer = NULL;
-    proc->state = PROC_STOPPED;
+void
+sched_yieldk()
+{
+    cpu_enable_interrupt();
+    cpu_int(LUNAIX_SCHED);

 }
 
 }
 

-__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……
+__DEFINE_LXSYSCALL1(unsigned int, sleep, unsigned int, seconds)
+{

     if (!seconds) {
         return 0;
     }
     if (!seconds) {
         return 0;
     }

-    if (__current->timer) {
-        return __current->timer->counter / timer_context()->running_frequency;
+
+    if (__current->sleep.wakeup_time) {
+        return (__current->sleep.wakeup_time - clock_systime()) / 1000U;

     }
 
     }
 

-    struct lx_timer* timer = timer_run_second(seconds, proc_timer_callback, __current, 0);
-    __current->timer = timer;
+    struct proc_info* root_proc = sched_ctx._procs[0];
+    __current->sleep.wakeup_time = clock_systime() + seconds * 1000;
+    llist_append(&root_proc->sleep.sleepers, &__current->sleep.sleepers);
+

     __current->intr_ctx.registers.eax = seconds;
     __current->intr_ctx.registers.eax = seconds;

-    __current->state = PROC_BLOCKED;
+    __current->state = PS_BLOCKED;

     schedule();
 }
 
     schedule();
 }
 

-__DEFINE_LXSYSCALL1(void, exit, int, status) {
+__DEFINE_LXSYSCALL1(unsigned int, alarm, unsigned int, seconds)
+{
+    time_t prev_ddl = __current->sleep.alarm_time;
+    time_t now = clock_systime();
+
+    __current->sleep.alarm_time = seconds ? now + seconds * 1000 : 0;
+
+    struct proc_info* root_proc = sched_ctx._procs[0];
+    if (llist_empty(&__current->sleep.sleepers)) {
+        llist_append(&root_proc->sleep.sleepers, &__current->sleep.sleepers);
+    }
+
+    return prev_ddl ? (prev_ddl - now) / 1000 : 0;
+}
+
+__DEFINE_LXSYSCALL1(void, exit, int, status)
+{

     terminate_proc(status);
     terminate_proc(status);

+    schedule();

 }
 
 }
 

-__DEFINE_LXSYSCALL(void, yield) {
+__DEFINE_LXSYSCALL(void, yield)
+{

     schedule();
 }
 
     schedule();
 }
 

-__DEFINE_LXSYSCALL1(pid_t, wait, int*, status) {
+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);
+}
+
+__DEFINE_LXSYSCALL(int, geterrno)
+{
+    return __current->k_status;
+}
+
+pid_t
+_wait(pid_t wpid, int* status, int options)
+{

     pid_t cur = __current->pid;
     pid_t cur = __current->pid;

+    int status_flags = 0;

     struct proc_info *proc, *n;
     if (llist_empty(&__current->children)) {
         return -1;
     }
     struct proc_info *proc, *n;
     if (llist_empty(&__current->children)) {
         return -1;
     }

+
+    wpid = wpid ? wpid : -__current->pgid;

 repeat:
 repeat:

-    llist_for_each(proc, n, &__current->children, siblings) {
-        if (proc->state == PROC_TERMNAT) {
-            goto done;
+    llist_for_each(proc, n, &__current->children, siblings)
+    {
+        if (!~wpid || proc->pid == wpid || proc->pgid == -wpid) {
+            if (proc->state == PS_TERMNAT && !options) {
+                status_flags |= PEXITTERM;
+                goto done;
+            }
+            if (proc->state == PS_READY && (options & WUNTRACED)) {
+                status_flags |= PEXITSTOP;
+                goto done;
+            }

         }
     }
         }
     }

-    // FIXME: 除了循环，也许有更高效的办法…… (在这里进行schedule，需要重写context switch!)
+    if ((options & WNOHANG)) {
+        return 0;
+    }
+    // 放弃当前的运行机会
+    sched_yieldk();

     goto repeat;
 
 done:
     goto repeat;
 
 done:

-    *status = proc->exit_code;
+    status_flags |= PEXITSIG * (proc->sig_inprogress != 0);
+    if (status) {
+        *status = proc->exit_code | status_flags;
+    }

     return destroy_process(proc->pid);
 }
 
     return destroy_process(proc->pid);
 }
 

-pid_t alloc_pid() {
+struct proc_info*
+alloc_process()
+{

     pid_t i = 0;
     pid_t i = 0;

-    for (; i < sched_ctx.ptable_len && sched_ctx._procs[i].state != PROC_DESTROY; i++);
+    for (; i < sched_ctx.ptable_len && sched_ctx._procs[i]; i++)
+        ;

 
     if (i == MAX_PROCESS) {
 
     if (i == MAX_PROCESS) {

-        panick("Process table is full");
+        panick("Panic in Ponyville shimmer!");

     }
     }

-    return i + 1;
-}

 
 

-void push_process(struct proc_info* process) {
-    int index = process->pid - 1;
-    if (index < 0 || index > sched_ctx.ptable_len) {
-        __current->k_status = LXINVLDPID;
-        return;
-    }
-    
-    if (index == sched_ctx.ptable_len) {
+    if (i == sched_ctx.ptable_len) {

         sched_ctx.ptable_len++;
     }
         sched_ctx.ptable_len++;
     }

-    
-    sched_ctx._procs[index] = *process;

 
 

-    process = &sched_ctx._procs[index];
+    struct proc_info* proc = cake_grab(proc_pile);

 
 

-    // make sure the address is in the range of process table
-    llist_init_head(&process->children);
-    // every process is the child of first process (pid=1)
-    if (process->parent) {
-        llist_append(&process->parent->children, &process->siblings);
+    proc->state = PS_CREATED;
+    proc->pid = i;
+    proc->created = clock_systime();
+    proc->pgid = proc->pid;
+    proc->fdtable = vzalloc(sizeof(struct v_fdtable));
+
+    llist_init_head(&proc->mm.regions.head);
+    llist_init_head(&proc->children);
+    llist_init_head(&proc->grp_member);
+    llist_init_head(&proc->sleep.sleepers);
+    waitq_init(&proc->waitqueue);
+
+    sched_ctx._procs[i] = proc;
+
+    return proc;
+}
+
+void
+commit_process(struct proc_info* process)
+{
+    assert(process == sched_ctx._procs[process->pid]);
+
+    if (process->state != PS_CREATED) {
+        __current->k_status = EINVAL;
+        return;

     }
     }

-    else {
-        process->parent = &sched_ctx._procs[0];
+
+    // every process is the child of first process (pid=1)
+    if (!process->parent) {
+        process->parent = sched_ctx._procs[1];

     }
 
     }
 

-    process->state = PROC_STOPPED;    
+    llist_append(&process->parent->children, &process->siblings);
+
+    process->state = PS_READY;

 }
 
 // from <kernel/process.c>
 }
 
 // from <kernel/process.c>

-extern void __del_pagetable(pid_t pid, uintptr_t mount_point);
+extern void
+__del_pagetable(pid_t pid, uintptr_t mount_point);

 
 

-pid_t destroy_process(pid_t pid) {
-    int index = pid - 1;
+pid_t
+destroy_process(pid_t pid)
+{
+    int index = pid;

     if (index <= 0 || index > sched_ctx.ptable_len) {
     if (index <= 0 || index > sched_ctx.ptable_len) {

-        __current->k_status = LXINVLDPID;
+        __current->k_status = EINVAL;

         return;
     }
         return;
     }

-    struct proc_info *proc = &sched_ctx._procs[index];
-    proc->state = PROC_DESTROY;
+    struct proc_info* proc = sched_ctx._procs[index];
+    sched_ctx._procs[index] = 0;
+

     llist_delete(&proc->siblings);
 
     llist_delete(&proc->siblings);
 

-    if (proc->mm.regions) {
-        struct mm_region *pos, *n;
-        llist_for_each(pos, n, &proc->mm.regions->head, head) {
-            lxfree(pos);
-        }
+    for (size_t i = 0; i < VFS_MAX_FD; i++) {
+        struct v_fd* fd = proc->fdtable->fds[i];
+        if (fd)
+            vfs_close(fd->file);
+    }
+
+    vfree(proc->fdtable);
+
+    struct mm_region *pos, *n;
+    llist_for_each(pos, n, &proc->mm.regions.head, head)
+    {
+        vfree(pos);

     }
 
     }
 

-    vmm_mount_pd(PD_MOUNT_2, proc->page_table);
+    vmm_mount_pd(PD_MOUNT_1, proc->page_table);

 
 

-    __del_pagetable(pid, PD_MOUNT_2);
+    __del_pagetable(pid, PD_MOUNT_1);

 
 

-    vmm_unmount_pd(PD_MOUNT_2);
+    vmm_unmount_pd(PD_MOUNT_1);
+
+    cake_release(proc_pile, proc);

 
     return pid;
 }
 
 
     return pid;
 }
 

-void terminate_proc(int exit_code) {
-    __current->state = PROC_TERMNAT;
+void
+terminate_proc(int exit_code)
+{
+    __current->state = PS_TERMNAT;

     __current->exit_code = exit_code;
 
     __current->exit_code = exit_code;
 

-    schedule();
+    __SIGSET(__current->parent->sig_pending, _SIGCHLD);

 }
 
 }
 

-struct proc_info* get_process(pid_t pid) {
-    int index = pid - 1;
+struct proc_info*
+get_process(pid_t pid)
+{
+    int index = pid;

     if (index < 0 || index > sched_ctx.ptable_len) {
         return NULL;
     }
     if (index < 0 || index > sched_ctx.ptable_len) {
         return NULL;
     }

-    return &sched_ctx._procs[index];
+    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-1];
+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;
     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
     // 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;
+    return PROC_TERMINATED(parent->state) || parent->created > proc->created;

 }
\ No newline at end of file
 }
\ No newline at end of file