sched/tcb: add a reference count to the TCB to prevent it from being deleted.

To replace the large lock with smaller ones and reduce the large locks related to the TCB,
in many scenarios, we only need to ensure that the TCB won't be released
instead of locking, thus reducing the possibility of lock recursion.

Signed-off-by: hujun5 <hujun5@xiaomi.com>

Author: hujun260

arch/risc-v/src/common/riscv_exception.c

@@ -111,6 +111,7 @@ int riscv_exception(int mcause, void *regs, void *args)
       _alert("Segmentation fault in %s (PID %d: %s)\n", get_task_name(ptcb),
              tcb->pid, get_task_name(tcb));
 
+      nxsched_put_tcb(ptcb);
       tcb->flags |= TCB_FLAG_FORCED_CANCEL;
 
       /* Return to _exit function in privileged mode with argument SIGSEGV */

drivers/note/note_driver.c

@@ -2109,6 +2109,7 @@ FAR char *note_get_taskname(pid_t pid, FAR char *buf, size_t len)
   if (tcb != NULL)
     {
       strlcpy(buf, tcb->name, len);
+      nxsched_put_tcb(tcb);
       return buf;
     }
 

drivers/syslog/vsyslog.c

@@ -36,6 +36,7 @@
 #include <nuttx/streams.h>
 #include <nuttx/syslog/syslog.h>
 
+#include "sched/sched.h"
 #include "syslog.h"
 
 /****************************************************************************

fs/inode/fs_files.c

@@ -255,6 +255,7 @@ static void task_fssync(FAR struct tcb_s *tcb, FAR void *arg)
           ctcb = nxsched_get_tcb(pid);
           if (ctcb == NULL || ctcb->group == NULL || ctcb != tcb)
             {
+              nxsched_put_tcb(ctcb);
               return;
             }
 
@@ -264,6 +265,8 @@ static void task_fssync(FAR struct tcb_s *tcb, FAR void *arg)
               file_fsync(filep);
               file_put(filep);
             }
+
+          nxsched_put_tcb(ctcb);
         }
     }
 }

fs/procfs/fs_procfs.c

@@ -942,8 +942,7 @@ static int procfs_readdir(FAR struct inode *mountpt,
           /* Verify that the pid still refers to an active task/thread */
 
           pid_t pid = level0->pid[index];
-          FAR struct tcb_s *tcb = nxsched_get_tcb(pid);
-          if (!tcb)
+          if (!nxsched_verify_pid(pid))
             {
               ferr("ERROR: PID %d is no longer valid\n", pid);
               return -ENOENT;

fs/procfs/fs_procfsmeminfo.c

@@ -537,6 +537,7 @@ static ssize_t memdump_write(FAR struct file *filep, FAR const char *buffer,
         }
 
       tcb->flags |= TCB_FLAG_HEAP_DUMP;
+      nxsched_put_tcb(tcb);
       return buflen;
     }
   else if ((p = strstr(buffer, "off")) != NULL)
@@ -550,6 +551,7 @@ static ssize_t memdump_write(FAR struct file *filep, FAR const char *buffer,
         }
 
       tcb->flags &= ~TCB_FLAG_HEAP_DUMP;
+      nxsched_put_tcb(tcb);
       return buflen;
     }
 #endif

fs/procfs/fs_procfsproc.c

@@ -62,6 +62,7 @@
 #include <nuttx/lib/lib.h>
 
 #include "fs_heap.h"
+#include "sched/sched.h"
 
 #if !defined(CONFIG_SCHED_CPULOAD_NONE) || defined(CONFIG_SCHED_CRITMONITOR)
 #  include <nuttx/clock.h>
@@ -1462,7 +1463,6 @@ static int proc_open(FAR struct file *filep, FAR const char *relpath,
 {
   FAR struct proc_file_s *procfile;
   FAR const struct proc_node_s *node;
-  FAR struct tcb_s *tcb;
   FAR char *ptr;
   unsigned long tmp;
   pid_t pid;
@@ -1508,9 +1508,7 @@ static int proc_open(FAR struct file *filep, FAR const char *relpath,
   /* Now verify that a task with this task/thread ID exists */
 
   pid = (pid_t)tmp;
-
-  tcb = nxsched_get_tcb(pid);
-  if (tcb == NULL)
+  if (!nxsched_verify_pid(pid))
     {
       ferr("ERROR: PID %d is no longer valid\n", pid);
       return -ENOENT;
@@ -1663,6 +1661,8 @@ static ssize_t proc_read(FAR struct file *filep, FAR char *buffer,
 
   leave_critical_section(flags);
 
+  nxsched_put_tcb(tcb);
+
   /* Update the file offset */
 
   if (ret > 0)
@@ -1714,6 +1714,7 @@ static ssize_t proc_write(FAR struct file *filep, FAR const char *buffer,
         break;
     }
 
+  nxsched_put_tcb(tcb);
   return ret;
 }
 
@@ -1769,7 +1770,6 @@ static int proc_opendir(FAR const char *relpath,
 {
   FAR struct proc_dir_s *procdir;
   FAR const struct proc_node_s *node;
-  FAR struct tcb_s *tcb;
   unsigned long tmp;
   FAR char *ptr;
   pid_t pid;
@@ -1819,9 +1819,7 @@ static int proc_opendir(FAR const char *relpath,
   /* Now verify that a task with this task/thread ID exists */
 
   pid = (pid_t)tmp;
-
-  tcb = nxsched_get_tcb(pid);
-  if (tcb == NULL)
+  if (!nxsched_verify_pid(pid))
     {
       ferr("ERROR: PID %d is not valid\n", pid);
       return -ENOENT;
@@ -1911,7 +1909,6 @@ static int proc_readdir(FAR struct fs_dirent_s *dir,
 {
   FAR struct proc_dir_s *procdir;
   FAR const struct proc_node_s *node = NULL;
-  FAR struct tcb_s *tcb;
   unsigned int index;
   pid_t pid;
   int ret;
@@ -1939,9 +1936,7 @@ static int proc_readdir(FAR struct fs_dirent_s *dir,
       /* Verify that the pid still refers to an active task/thread */
 
       pid = procdir->pid;
-
-      tcb = nxsched_get_tcb(pid);
-      if (tcb == NULL)
+      if (!nxsched_verify_pid(pid))
         {
           ferr("ERROR: PID %d is no longer valid\n", pid);
           return -ENOENT;
@@ -2013,7 +2008,6 @@ static int proc_rewinddir(struct fs_dirent_s *dir)
 static int proc_stat(const char *relpath, struct stat *buf)
 {
   FAR const struct proc_node_s *node;
-  FAR struct tcb_s *tcb;
   unsigned long tmp;
   FAR char *ptr;
   pid_t pid;
@@ -2057,9 +2051,7 @@ static int proc_stat(const char *relpath, struct stat *buf)
   /* Now verify that a task with this task/thread ID exists */
 
   pid = (pid_t)tmp;
-
-  tcb = nxsched_get_tcb(pid);
-  if (tcb == NULL)
+  if (!nxsched_verify_pid(pid))
     {
       ferr("ERROR: PID %d is no longer valid\n", pid);
       return -ENOENT;

include/nuttx/mm/mm.h

@@ -148,7 +148,7 @@
     ((node) != NULL && (dump)->pid == (node)->pid)
 #  define MM_DUMP_LEAK(dump, node) \
     ((node) != NULL && (dump)->pid == PID_MM_LEAK && (node)->pid >= 0 && \
-     nxsched_get_tcb((node)->pid) == NULL)
+    !nxsched_verify_pid((node)->pid))
 #else
 #  define MM_DUMP_ALLOC(dump,node)  ((dump)->pid == PID_MM_ALLOC)
 #  define MM_DUMP_SEQNO(dump,node)  (true)

include/nuttx/sched.h

@@ -111,6 +111,7 @@
 #define TCB_FLAG_JOIN_COMPLETED    (1 << 14)                     /* Bit 14: Pthread join completed */
 #define TCB_FLAG_FREE_TCB          (1 << 15)                     /* Bit 15: Free tcb after exit */
 #define TCB_FLAG_PREEMPT_SCHED     (1 << 16)                     /* Bit 16: tcb is PREEMPT_SCHED */
+#define TCB_FLAG_KILL_PROCESSING   (1 << 17)                     /* Bit 17: tcb is killed */
 
 /* Values for struct task_group tg_flags */
 
@@ -727,6 +728,16 @@ struct tcb_s
   size_t level_deepest;
   size_t level;
 #endif
+
+  /* The total number that we are referenced by other tasks and
+   * The total number of other tasks that we referenced.
+   */
+
+  atomic_t refs;
+
+  /* When we exit, we need post this sem. */
+
+  sem_t    exit_sem;
 };
 
 /* struct task_tcb_s ********************************************************/
@@ -904,25 +915,21 @@ FAR struct tcb_s *nxsched_self(void);
 void nxsched_foreach(nxsched_foreach_t handler, FAR void *arg);
 
 /****************************************************************************
- * Name: nxsched_get_tcb
+ * Name: nxsched_get_tcb/nxsched_put_tcb
  *
  * Description:
- *   Given a task ID, this function will return the a pointer to the
- *   corresponding TCB (or NULL if there is no such task ID).
- *
- *   NOTE:  This function holds a critical section while examining TCB data
- *   data structures but releases that critical section before returning.
- *   When it is released, the TCB may become unstable.  If the caller
- *   requires absolute stability while using the TCB, then the caller
- *   should establish the critical section BEFORE calling this function and
- *   hold that critical section as long as necessary.
+ *   Given a task ID,
+ *   Obtain a valid TCB and increment the corresponding reference count to
+ *   prevent it from being released. nxsched_get_tcb and nxsched_put_tcb
+ *   must be called in pairs to ensure the proper release of the TCB.
  *
  ****************************************************************************/
 
 FAR struct tcb_s *nxsched_get_tcb(pid_t pid);
+void nxsched_put_tcb(FAR struct tcb_s *tcb);
 
 /****************************************************************************
- * Name:  nxsched_releasepid
+ * Name:  nxsched_release_tcb
  *
  * Description:
  *   When a task is destroyed, this function must be called to make its

libs/libc/gdbstub/lib_gdbstub.c

@@ -1028,6 +1028,8 @@ static void gdb_get_registers(FAR struct gdb_state_s *state)
             *(FAR uintptr_t *)(reg + g_tcbinfo.reg_off.p[i]);
         }
     }
+
+  nxsched_put_tcb(tcb);
 }
 
 /****************************************************************************
@@ -1424,6 +1426,7 @@ static int gdb_query(FAR struct gdb_state_s *state)
                 get_task_name(tcb), thread_state, tcb->sched_priority,
                 tcb->adj_stack_size);
 
+      nxsched_put_tcb(tcb);
       ret = gdb_bin2hex(state->pkt_buf, sizeof(state->pkt_buf),
                         thread_info, strlen(thread_info));
 
@@ -1502,6 +1505,7 @@ static int gdb_is_thread_active(FAR struct gdb_state_s *state)
 
   state->pid = pid - 1;
   gdb_send_ok_packet(state);
+  nxsched_put_tcb(tcb);
   return ret;
 }
 
@@ -1558,6 +1562,7 @@ static int gdb_thread_context(FAR struct gdb_state_s *state)
         }
 
       state->pid = pid - 1;
+      nxsched_put_tcb(tcb);
     }
 
   gdb_send_ok_packet(state);