refactor: Optimize the signal context overhead

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

diff --git a/lunaix-os/kernel/process/sched.c b/lunaix-os/kernel/process/sched.c
index ade13b31c6414f4e1e1588b26645f255475f2c6c..acf04b72111d6136cf77fbcc2a0e2fa329346bc9 100644 (file)
--- a/lunaix-os/kernel/process/sched.c
+++ b/lunaix-os/kernel/process/sched.c
@@ -1,5 +1,5 @@
+#include <arch/abi.h>
 #include <arch/x86/interrupts.h>
-#include <arch/x86/tss.h>
 
 #include <hal/apic.h>
 #include <hal/cpu.h>
@@ -63,20 +63,24 @@ sched_init_dummy()
     struct exec_param* execp =
       (void*)dummy_stack + DUMMY_STACK_SIZE - sizeof(struct exec_param);
 
+    isr_param* isrp = (void*)execp - sizeof(isr_param);
+
     *execp = (struct exec_param){
         .cs = KCODE_SEG,
         .eflags = cpu_reflags() | 0x0200,
-        .eip = (void*)my_dummy,
+        .eip = (ptr_t)my_dummy,
         .ss = KDATA_SEG,
     };
 
+    *isrp = (isr_param){ .registers = { .ds = KDATA_SEG,
+                                        .es = KDATA_SEG,
+                                        .fs = KDATA_SEG,
+                                        .gs = KDATA_SEG },
+                         .execp = execp };
+
     // memset to 0
     dummy_proc = (struct proc_info){};
-    dummy_proc.intr_ctx = (isr_param){ .registers = { .ds = KDATA_SEG,
-                                                      .es = KDATA_SEG,
-                                                      .fs = KDATA_SEG,
-                                                      .gs = KDATA_SEG },
-                                       .execp = execp };
+    dummy_proc.intr_ctx = isrp;
 
     dummy_proc.page_table = cpu_rcr3();
     dummy_proc.state = PS_READY;
@@ -99,7 +103,6 @@ run(struct proc_info* proc)
         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.esp);
 
     apic_done_servicing();
 
@@ -111,14 +114,24 @@ run(struct proc_info* proc)
 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)) {
+    if (!proc) {
+        return 0;
+    }
+
+    struct sighail* sh = &proc->sigctx;
+
+    if ((proc->state & PS_PAUSED)) {
+        return !!(sh->sig_pending & ~1);
+    }
+
+    if (sigset_test(sh->sig_pending, _SIGCONT)) {
+        sigset_clear(sh->sig_pending, _SIGSTOP);
+    } else if (sigset_test(sh->sig_pending, _SIGSTOP)) {
         // 如果进程受到SIGSTOP，则该进程不给予调度。
         return 0;
     }
 
-    return 1;
+    return (proc->state == PS_READY);
 }
 
 void
@@ -129,7 +142,7 @@ check_sleepers()
     time_t now = clock_systime();
     llist_for_each(pos, n, &leader->sleep.sleepers, sleep.sleepers)
     {
-        if (PROC_TERMINATED(pos->state)) {
+        if (proc_terminated(pos)) {
             goto del;
         }
 
@@ -143,7 +156,7 @@ check_sleepers()
 
         if (atime && now >= atime) {
             pos->sleep.alarm_time = 0;
-            __SIGSET(pos->sig_pending, _SIGALRM);
+            proc_setsignal(pos, _SIGALRM);
         }
 
         if (!wtime && !atime) {
@@ -165,6 +178,7 @@ schedule()
     struct proc_info* next;
     int prev_ptr = sched_ctx.procs_index;
     int ptr = prev_ptr;
+    int found = 0;
 
     if (!(__current->state & ~PS_RUNNING)) {
         __current->state = PS_READY;
@@ -173,24 +187,19 @@ schedule()
     check_sleepers();
 
     // round-robin scheduler
-redo:
     do {
         ptr = (ptr + 1) % sched_ctx.ptable_len;
         next = sched_ctx._procs[ptr];
-    } while (!next || (next->state != PS_READY && ptr != prev_ptr));
-
-    sched_ctx.procs_index = ptr;
 
-    if (next->state != PS_READY) {
-        // schedule the dummy process if we're out of choice
-        next = &dummy_proc;
-        goto done;
-    }
+        if (!(found = can_schedule(next))) {
+            if (ptr == prev_ptr) {
+                next = &dummy_proc;
+                goto done;
+            }
+        }
+    } while (!found);
 
-    if (!can_schedule(next)) {
-        // 如果该进程不给予调度，则尝试重新选择
-        goto redo;
-    }
+    sched_ctx.procs_index = ptr;
 
 done:
     run(next);
@@ -215,12 +224,17 @@ __DEFINE_LXSYSCALL1(unsigned int, sleep, unsigned int, seconds)
 
     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;
+    if (llist_empty(&__current->sleep.sleepers)) {
+        llist_append(&root_proc->sleep.sleepers, &__current->sleep.sleepers);
+    }
+
+    store_retval(seconds);
 
     block_current();
     schedule();
+
+    return 0;
 }
 
 __DEFINE_LXSYSCALL1(unsigned int, alarm, unsigned int, seconds)
@@ -300,7 +314,6 @@ repeat:
     goto repeat;
 
 done:
-    status_flags |= PEXITSIG * (proc->sig_inprogress != 0);
     if (status) {
         *status = proc->exit_code | status_flags;
     }
@@ -330,8 +343,6 @@ alloc_process()
     proc->created = clock_systime();
     proc->pgid = proc->pid;
     proc->fdtable = vzalloc(sizeof(struct v_fdtable));
-    proc->fxstate =
-      vzalloc_dma(512); // FXSAVE需要十六位对齐地址，使用DMA块（128位对齐）
 
     llist_init_head(&proc->mm.regions);
     llist_init_head(&proc->tasks);
@@ -368,7 +379,7 @@ commit_process(struct proc_info* process)
 
 // from <kernel/process.c>
 extern void
-__del_pagetable(pid_t pid, uintptr_t mount_point);
+__del_pagetable(pid_t pid, ptr_t mount_point);
 
 pid_t
 destroy_process(pid_t pid)
@@ -376,8 +387,9 @@ destroy_process(pid_t pid)
     int index = pid;
     if (index <= 0 || index > sched_ctx.ptable_len) {
         __current->k_status = EINVAL;
-        return;
+        return -1;
     }
+
     struct proc_info* proc = sched_ctx._procs[index];
     sched_ctx._procs[index] = 0;
 
@@ -401,7 +413,6 @@ destroy_process(pid_t pid)
     }
 
     vfree(proc->fdtable);
-    vfree_dma(proc->fxstate);
 
     vmm_mount_pd(VMS_MOUNT_1, proc->page_table);
 
@@ -427,7 +438,7 @@ terminate_proc(int exit_code)
     __current->state = PS_TERMNAT;
     __current->exit_code = exit_code;
 
-    __SIGSET(__current->parent->sig_pending, _SIGCHLD);
+    proc_setsignal(__current->parent, _SIGCHLD);
 }
 
 struct proc_info*
@@ -452,5 +463,5 @@ 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
-    return PROC_TERMINATED(parent->state) || parent->created > proc->created;
+    return proc_terminated(parent) || parent->created > proc->created;
 }
\ No newline at end of file