feat: lru eviction policy on page caches

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

diff --git a/lunaix-os/kernel/sched.c b/lunaix-os/kernel/sched.c
index 026fb515a8dbfbc34410e868de18ccad40b7b308..f8f00d40a094d62c31976cd4d54ef577673ba6de 100644 (file)
--- a/lunaix-os/kernel/sched.c
+++ b/lunaix-os/kernel/sched.c
@@ -5,6 +5,8 @@
 #include <hal/cpu.h>
 
 #include <lunaix/mm/kalloc.h>
 #include <hal/cpu.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/mm/vmm.h>
 #include <lunaix/process.h>
 #include <lunaix/sched.h>


@@ -20,8 +22,6 @@ 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")


@@ -30,11 +30,14 @@ void
 sched_init()
 {
     size_t pg_size = ROUNDUP(sizeof(struct proc_info) * MAX_PROCESS, 0x1000);
 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,
+    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 };
 }
                                     .ptable_len = 0,
                                     .procs_index = 0 };
 }


@@ -42,29 +45,68 @@ sched_init()
 void
 run(struct proc_info* proc)
 {
 void
 run(struct proc_info* proc)
 {

-    if (!(__current->state & ~PROC_RUNNING)) {
-        __current->state = PROC_STOPPED;
-    }
-    proc->state = PROC_RUNNING;
+    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);

 
 

-    // FIXME: 这里还是得再考虑一下。
-    // tss_update_esp(__current->intr_ctx.esp);
+    apic_done_servicing();

 
 

-    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;
+    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;
+        }

 
 

-    signal_dispatch();
+        time_t wtime = pos->sleep.wakeup_time;
+        time_t atime = pos->sleep.alarm_time;

 
 

-    asm volatile("movl %0, %%eax\n"
-                 "jmp soft_iret\n" ::"r"(&__current->intr_ctx)
-                 : "eax", "memory");
+        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
 }
 
 void


@@ -79,46 +121,74 @@ schedule()
     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)
+void
+sched_yieldk()

 {
 {

-    proc->timer = NULL;
-    proc->state = PROC_STOPPED;
+    cpu_int(LUNAIX_SCHED);

 }
 
 __DEFINE_LXSYSCALL1(unsigned int, sleep, unsigned int, seconds)
 {
 }
 
 __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 (!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(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);
 __DEFINE_LXSYSCALL1(void, exit, int, status)
 {
     terminate_proc(status);

+    schedule();

 }
 
 __DEFINE_LXSYSCALL(void, yield)
 }
 
 __DEFINE_LXSYSCALL(void, yield)


@@ -139,6 +209,11 @@ __DEFINE_LXSYSCALL3(pid_t, waitpid, pid_t, pid, int*, status, int, options)
     return _wait(pid, status, 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
 _wait(pid_t wpid, int* status, int options)
 {


@@ -150,17 +225,16 @@ _wait(pid_t wpid, int* status, int options)
     }
 
     wpid = wpid ? wpid : -__current->pgid;
     }
 
     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) {
 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;
+            if (proc->state == PS_TERMNAT && !options) {
+                status_flags |= PEXITTERM;

                 goto done;
             }
                 goto done;
             }

-            if (proc->state == PROC_STOPPED && (options & WUNTRACED)) {
-                status_flags |= PROCSTOP;
+            if (proc->state == PS_STOPPED && (options & WUNTRACED)) {
+                status_flags |= PEXITSTOP;

                 goto done;
             }
         }
                 goto done;
             }
         }


@@ -169,59 +243,68 @@ repeat:
         return 0;
     }
     // 放弃当前的运行机会
         return 0;
     }
     // 放弃当前的运行机会

-    sched_yield();
+    sched_yieldk();

     goto repeat;
 
 done:
     goto repeat;
 
 done:

-    cpu_disable_interrupt();
-    *status = (proc->exit_code & 0xffff) | status_flags;
+    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;
+    for (; i < sched_ctx.ptable_len && sched_ctx._procs[i].state != PS_DESTROY;

          i++)
         ;
 
     if (i == MAX_PROCESS) {
         panick("Panic in Ponyville shimmer!");
     }
          i++)
         ;
 
     if (i == MAX_PROCESS) {
         panick("Panic in Ponyville shimmer!");
     }

-    return i;
-}
-
-void
-push_process(struct proc_info* process)
-{
-    int index = process->pid;
-    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;
+    struct proc_info* proc = &sched_ctx._procs[i];
+    memset(proc, 0, sizeof(*proc));

 
 

-    process = &sched_ctx._procs[index];
+    proc->state = PS_CREATED;
+    proc->pid = i;
+    proc->created = clock_systime();
+    proc->pgid = proc->pid;
+    proc->fdtable = vzalloc(sizeof(struct v_fdtable));

 
 

-    // make sure the reference is relative to process table
-    llist_init_head(&process->children);
-    llist_init_head(&process->grp_member);
+    llist_init_head(&proc->mm.regions.head);
+    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 = EINVAL;
+        return;
+    }

 
     // every process is the child of first process (pid=1)
 
     // 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];
+    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>


@@ -233,26 +316,32 @@ destroy_process(pid_t pid)
 {
     int index = pid;
     if (index <= 0 || index > sched_ctx.ptable_len) {
 {
     int index = pid;
     if (index <= 0 || index > sched_ctx.ptable_len) {

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

         return;
     }
     struct proc_info* proc = &sched_ctx._procs[index];
         return;
     }
     struct proc_info* proc = &sched_ctx._procs[index];

-    proc->state = PROC_DESTROY;
+    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);
-        }
+    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);

 
     return pid;
 }
 
     return pid;
 }


@@ -260,10 +349,10 @@ destroy_process(pid_t pid)
 void
 terminate_proc(int exit_code)
 {
 void
 terminate_proc(int exit_code)
 {

-    __current->state = PROC_TERMNAT;
+    __current->state = PS_TERMNAT;

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

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

 }
 
 struct proc_info*
 }
 
 struct proc_info*


@@ -288,5 +377,5 @@ orphaned_proc(pid_t pid)
 
     // 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