git://scm.lunaixsky.com / lunaix-os.git / blobdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | inline | side by side
fix: race condition and partial state issue on injecting signal context into user...
[lunaix-os.git] / lunaix-os / kernel / sched.c
diff --git a/lunaix-os/kernel/sched.c b/lunaix-os/kernel/sched.c
index 758b157befb68a2b48656ab90f9887067c0f4997..8e089d64b976be61ddf22fb77259caf9dea54f5b 100644 (file)
--- a/lunaix-os/kernel/sched.c
+++ b/lunaix-os/kernel/sched.c
@@ -1,16 +1,19 @@
-#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/kalloc.h>
+#include <lunaix/mm/pmm.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 MAX_PROCESS 512
 
@@ -18,209 +21,342 @@ volatile struct proc_info* __current;
 
 struct proc_info dummy;
 
 
 struct proc_info dummy;
 
-extern void __proc_table;
-
 struct scheduler sched_ctx;
 
 LOG_MODULE("SCHED")
 
 struct scheduler sched_ctx;
 
 LOG_MODULE("SCHED")
 
-void sched_init() {
+void
+sched_init()
+{
     size_t pg_size = ROUNDUP(sizeof(struct proc_info) * MAX_PROCESS, 0x1000);
     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 run(struct proc_info* proc) {
-    if (!(__current->state & ~PROC_RUNNING)) {
-        __current->state = PROC_STOPPED;
+    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->state = PROC_RUNNING;
-    
-    // FIXME: 这里还是得再考虑一下。
-    // tss_update_esp(__current->intr_ctx.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;
-    }
+    sched_ctx = (struct scheduler){ ._procs = (struct proc_info*)PROC_START,
+                                    .ptable_len = 0,
+                                    .procs_index = 0 };
+}
+
+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);
 
     apic_done_servicing();
 
 
     apic_done_servicing();
 
-    asm volatile (
-        "pushl %0\n"
-        "jmp soft_iret\n"::"r"(&__current->intr_ctx): "memory");
+    asm volatile("pushl %0\n"
+                 "jmp switch_to\n" ::"r"(proc)
+                 : "memory"); // kernel/asm/x86/interrupt.S
 }
 
 }
 
-void schedule() {
+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;
+    }
+
+    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_STOPPED;
+        }
+
+        if (atime && now >= atime) {
+            pos->sleep.alarm_time = 0;
+            __SIGSET(pos->sig_pending, _SIGALRM);
+        }
+
+        if (!wtime && !atime) {
+        del:
+            llist_delete(&pos->sleep.sleepers);
+        }
+    }
+}
+
+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_STOPPED;
+    }
+
+    check_sleepers();
+
     // round-robin scheduler
     // round-robin scheduler
+redo:
     do {
         ptr = (ptr + 1) % sched_ctx.ptable_len;
         next = &sched_ctx._procs[ptr];
     do {
         ptr = (ptr + 1) % sched_ctx.ptable_len;
         next = &sched_ctx._procs[ptr];
-    } while(next->state != PROC_STOPPED && ptr != prev_ptr);
-    
+    } while (next->state != PS_STOPPED && 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;
-}
-
-__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;
+    __current->sleep.wakeup_time = clock_systime() + seconds * 1000;
+    llist_append(&sched_ctx._procs[0].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;
+
+    if (llist_empty(&__current->sleep.sleepers)) {
+        llist_append(&sched_ctx._procs[0].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);
+}
+
+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;
+    cpu_enable_interrupt();
 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_STOPPED && (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_yield();
     goto repeat;
 
 done:
     goto repeat;
 
 done:
-    *status = proc->exit_code;
+    cpu_disable_interrupt();
+    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].state != PS_DESTROY;
+         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 = &sched_ctx._procs[i];
+    memset(proc, 0, sizeof(*proc));
 
 
-    // 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;
+
+    llist_init_head(&proc->mm.regions);
+    llist_init_head(&proc->children);
+    llist_init_head(&proc->grp_member);
+    llist_init_head(&proc->sleep.sleepers);
+
+    return proc;
+}
+
+void
+commit_process(struct proc_info* process)
+{
+    assert(process == &sched_ctx._procs[process->pid]);
+
+    if (process->state != PS_CREATED) {
+        __current->k_status = LXINVL;
+        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_STOPPED;
 }
 
 // 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) {
         __current->k_status = LXINVLDPID;
         return;
     }
     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;
+    struct proc_info* proc = &sched_ctx._procs[index];
+    proc->state = PS_DESTROY;
     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);
-        }
+    struct mm_region *pos, *n;
+    llist_for_each(pos, n, &proc->mm.regions.head, head)
+    {
+        lxfree(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);
 
     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;
     }
     return &sched_ctx._procs[index];
 }
 
     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-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
The lunaix project - hobby kernel from scratch