diff --git a/include/nuttx/sched.h b/include/nuttx/sched.h
index 3ee50dc852..6458afa7ff 100644
--- a/include/nuttx/sched.h
+++ b/include/nuttx/sched.h
@@ -679,6 +679,12 @@ struct tcb_s
   FAR struct pthread_mutex_s *mhead;     /* List of mutexes held by thread  */
 #endif
 
+  /* CPU load monitoring support ********************************************/
+
+#ifdef CONFIG_SCHED_CPULOAD
+  uint32_t ticks;                        /* Number of ticks on this thread */
+#endif
+
   /* Pre-emption monitor support ********************************************/
 
 #ifdef CONFIG_SCHED_CRITMONITOR
diff --git a/sched/init/nx_start.c b/sched/init/nx_start.c
index f54095f3dc..0f034aeb22 100644
--- a/sched/init/nx_start.c
+++ b/sched/init/nx_start.c
@@ -197,8 +197,7 @@ volatile pid_t g_lastpid;
  * 2. Is used to quickly map a process ID into a TCB.
  */
 
-FAR struct pidhash_s *g_pidhash;
-
+FAR struct tcb_s **g_pidhash;
 volatile int g_npidhash;
 
 /* This is a table of task lists.  This table is indexed by the task stat
@@ -547,15 +546,9 @@ void nx_start(void)
       g_npidhash <<= 1;
     }
 
-  g_pidhash = kmm_malloc(sizeof(struct pidhash_s) * g_npidhash);
+  g_pidhash = kmm_zalloc(sizeof(*g_pidhash) * g_npidhash);
   DEBUGASSERT(g_pidhash);
 
-  for (i = 0; i < g_npidhash; i++)
-    {
-      g_pidhash[i].tcb = NULL;
-      g_pidhash[i].pid = INVALID_PROCESS_ID;
-    }
-
   /* IDLE Group Initialization **********************************************/
 
   for (i = 0; i < CONFIG_SMP_NCPUS; i++)
@@ -565,9 +558,8 @@ void nx_start(void)
 
       /* Assign the process ID(s) of ZERO to the idle task(s) */
 
-      hashndx                = PIDHASH(i);
-      g_pidhash[hashndx].tcb = &g_idletcb[i].cmn;
-      g_pidhash[hashndx].pid = i;
+      hashndx            = PIDHASH(i);
+      g_pidhash[hashndx] = &g_idletcb[i].cmn;
 
       /* Allocate the IDLE group */
 
@@ -815,7 +807,7 @@ void nx_start(void)
 
           flags = enter_critical_section();
 
-          tcb = g_pidhash[i].tcb;
+          tcb = g_pidhash[i];
           if (tcb && (up_check_tcbstack(tcb) * 100 / tcb->adj_stack_size
                       > CONFIG_STACK_USAGE_SAFE_PERCENT))
             {
diff --git a/sched/sched/sched.h b/sched/sched/sched.h
index b87c0107db..4543a2a1ae 100644
--- a/sched/sched/sched.h
+++ b/sched/sched/sched.h
@@ -90,26 +90,6 @@
  * Public Type Definitions
  ****************************************************************************/
 
-/* This structure defines the format of the hash table that is used to (1)
- * determine if a task ID is unique, and (2) to map a process ID to its
- * corresponding TCB.
- *
- * NOTE also that CPU load measurement data is retained in his table vs. in
- * the TCB which would seem to be the more logic place.  It is place in the
- * hash table, instead, to facilitate CPU load adjustments on all threads
- * during timer interrupt handling. nxsched_foreach() could do this too, but
- * this would require a little more overhead.
- */
-
-struct pidhash_s
-{
-  FAR struct tcb_s *tcb;       /* TCB assigned to this PID */
-  pid_t pid;                   /* The full PID value */
-#ifdef CONFIG_SCHED_CPULOAD
-  uint32_t ticks;              /* Number of ticks on this thread */
-#endif
-};
-
 /* This structure defines an element of the g_tasklisttable[].  This table
  * is used to map a task_state enumeration to the corresponding task list.
  */
@@ -243,8 +223,7 @@ extern volatile pid_t g_lastpid;
  * 2. Is used to quickly map a process ID into a TCB.
  */
 
-extern FAR struct pidhash_s *g_pidhash;
-
+extern FAR struct tcb_s **g_pidhash;
 extern volatile int g_npidhash;
 
 /* This is a table of task lists.  This table is indexed by the task stat
diff --git a/sched/sched/sched_cpuload.c b/sched/sched/sched_cpuload.c
index 5a16b70860..1f3fac3165 100644
--- a/sched/sched/sched_cpuload.c
+++ b/sched/sched/sched_cpuload.c
@@ -110,20 +110,11 @@ volatile uint32_t g_cpuload_total;
 
 static inline void nxsched_cpu_process_cpuload(int cpu)
 {
-  FAR struct tcb_s *rtcb  = current_task(cpu);
-  int hash_index;
+  FAR struct tcb_s *rtcb = current_task(cpu);
 
-  /* Increment the count on the currently executing thread
-   *
-   * NOTE also that CPU load measurement data is retained in the g_pidhash
-   * table vs. in the TCB which would seem to be the more logic place.  It
-   * is place in the hash table, instead, to facilitate CPU load adjustments
-   * on all threads during timer interrupt handling. nxsched_foreach() could
-   * do this too, but this would require a little more overhead.
-   */
+  /* Increment the count on the currently executing thread */
 
-  hash_index = PIDHASH(rtcb->pid);
-  g_pidhash[hash_index].ticks++;
+  rtcb->ticks++;
 
   /* Increment tick count.  NOTE that the count is increment once for each
    * CPU on each sample interval.
@@ -193,8 +184,8 @@ void weak_function nxsched_process_cpuload(void)
 
       for (i = 0; i < g_npidhash; i++)
         {
-          g_pidhash[i].ticks >>= 1;
-          total += g_pidhash[i].ticks;
+          g_pidhash[i]->ticks >>= 1;
+          total += g_pidhash[i]->ticks;
         }
 
       /* Save the new total. */
@@ -254,10 +245,10 @@ int clock_cpuload(int pid, FAR struct cpuload_s *cpuload)
    * do this too, but this would require a little more overhead.
    */
 
-  if (g_pidhash[hash_index].tcb && g_pidhash[hash_index].pid == pid)
+  if (g_pidhash[hash_index] && g_pidhash[hash_index]->pid == pid)
     {
       cpuload->total  = g_cpuload_total;
-      cpuload->active = g_pidhash[hash_index].ticks;
+      cpuload->active = g_pidhash[hash_index]->ticks;
       ret = OK;
     }
 
diff --git a/sched/sched/sched_foreach.c b/sched/sched/sched_foreach.c
index 0a074cb42b..fca65730ce 100644
--- a/sched/sched/sched_foreach.c
+++ b/sched/sched/sched_foreach.c
@@ -67,9 +67,9 @@ void nxsched_foreach(nxsched_foreach_t handler, FAR void *arg)
     {
       /* This test and the function call must be atomic */
 
-      if (g_pidhash[ndx].tcb)
+      if (g_pidhash[ndx])
         {
-          handler(g_pidhash[ndx].tcb, arg);
+          handler(g_pidhash[ndx], arg);
         }
     }
 
diff --git a/sched/sched/sched_gettcb.c b/sched/sched/sched_gettcb.c
index 373e21828f..9ee502cce3 100644
--- a/sched/sched/sched_gettcb.c
+++ b/sched/sched/sched_gettcb.c
@@ -76,11 +76,11 @@ FAR struct tcb_s *nxsched_get_tcb(pid_t pid)
 
       /* Verify that the correct TCB was found. */
 
-      if (g_pidhash && pid == g_pidhash[hash_ndx].pid)
+      if (g_pidhash && pid == g_pidhash[hash_ndx]->pid)
         {
           /* Return the TCB associated with this pid (if any) */
 
-          ret = g_pidhash[hash_ndx].tcb;
+          ret = g_pidhash[hash_ndx];
         }
 
       leave_critical_section(flags);
diff --git a/sched/sched/sched_releasetcb.c b/sched/sched/sched_releasetcb.c
index 1f95c4e1ba..12fdde5fba 100644
--- a/sched/sched/sched_releasetcb.c
+++ b/sched/sched/sched_releasetcb.c
@@ -51,23 +51,20 @@ static void nxsched_releasepid(pid_t pid)
   irqstate_t flags = enter_critical_section();
   int hash_ndx = PIDHASH(pid);
 
+#ifdef CONFIG_SCHED_CPULOAD
+  /* Decrement the total CPU load count held by this thread from the
+   * total for all threads.
+   */
+
+  g_cpuload_total -= g_pidhash[hash_ndx]->ticks;
+#endif
+
   /* Make any pid associated with this hash available.  Note:
    * no special precautions need be taken here because the
    * following action is atomic
    */
 
-  g_pidhash[hash_ndx].tcb   = NULL;
-  g_pidhash[hash_ndx].pid   = INVALID_PROCESS_ID;
-
-#ifdef CONFIG_SCHED_CPULOAD
-  /* Decrement the total CPU load count held by this thread from the
-   * total for all threads.  Then we can reset the count on this
-   * defunct thread to zero.
-   */
-
-  g_cpuload_total          -= g_pidhash[hash_ndx].ticks;
-  g_pidhash[hash_ndx].ticks = 0;
-#endif
+  g_pidhash[hash_ndx] = NULL;
 
   leave_critical_section(flags);
 }
diff --git a/sched/sched/sched_verifytcb.c b/sched/sched/sched_verifytcb.c
index 82db055e31..8703fd8c85 100644
--- a/sched/sched/sched_verifytcb.c
+++ b/sched/sched/sched_verifytcb.c
@@ -72,7 +72,7 @@ bool nxsched_verify_tcb(FAR struct tcb_s *tcb)
   bool vaild;
 
   flags = enter_critical_section();
-  vaild = tcb == g_pidhash[PIDHASH(tcb->pid)].tcb;
+  vaild = tcb == g_pidhash[PIDHASH(tcb->pid)];
   leave_critical_section(flags);
 
   return vaild;
diff --git a/sched/task/task_setup.c b/sched/task/task_setup.c
index c9e9eb73d7..7185d045dd 100644
--- a/sched/task/task_setup.c
+++ b/sched/task/task_setup.c
@@ -79,7 +79,7 @@ static const char g_noname[] = "<noname>";
 
 static int nxtask_assign_pid(FAR struct tcb_s *tcb)
 {
-  FAR struct pidhash_s *pidhash;
+  FAR struct tcb_s **pidhash;
   pid_t next_pid;
   int   hash_ndx;
   int   i;
@@ -117,15 +117,11 @@ retry:
 
       /* Check if there is a (potential) duplicate of this pid */
 
-      if (!g_pidhash[hash_ndx].tcb)
+      if (!g_pidhash[hash_ndx])
         {
           /* Assign this PID to the task */
 
-          g_pidhash[hash_ndx].tcb   = tcb;
-          g_pidhash[hash_ndx].pid   = next_pid;
-#ifdef CONFIG_SCHED_CPULOAD
-          g_pidhash[hash_ndx].ticks = 0;
-#endif
+          g_pidhash[hash_ndx] = tcb;
           tcb->pid = next_pid;
           g_lastpid = next_pid;
 
@@ -141,7 +137,7 @@ retry:
    * expand space.
    */
 
-  pidhash = kmm_malloc(g_npidhash * 2 * sizeof(struct pidhash_s));
+  pidhash = kmm_zalloc(g_npidhash * 2 * sizeof(*pidhash));
   if (pidhash == NULL)
     {
       leave_critical_section(flags);
@@ -150,24 +146,15 @@ retry:
 
   g_npidhash *= 2;
 
-  /* Reset the new hash table to the initial state */
-
-  for (i = 0; i < g_npidhash; i++)
-    {
-      pidhash[i].tcb = NULL;
-      pidhash[i].pid = INVALID_PROCESS_ID;
-    }
-
   /* All original pid and hash_ndx are mismatch,
    * so we need to rebuild their relationship
    */
 
   for (i = 0; i < g_npidhash / 2; i++)
     {
-      hash_ndx = PIDHASH(g_pidhash[i].pid);
-      DEBUGASSERT(pidhash[hash_ndx].tcb == NULL);
-      pidhash[hash_ndx].tcb = g_pidhash[i].tcb;
-      pidhash[hash_ndx].pid = g_pidhash[i].pid;
+      hash_ndx = PIDHASH(g_pidhash[i]->pid);
+      DEBUGASSERT(pidhash[hash_ndx] == NULL);
+      pidhash[hash_ndx] = g_pidhash[i];
     }
 
   /* Release resource for original g_pidhash, using new g_pidhash */
diff --git a/tools/esp32/backtrace.gdbscript b/tools/esp32/backtrace.gdbscript
index 79993ce492..41ae45011d 100644
--- a/tools/esp32/backtrace.gdbscript
+++ b/tools/esp32/backtrace.gdbscript
@@ -110,7 +110,7 @@ end
 
 define esp32_bt_pid
     set $_pid = $arg0
-    set $_tcb = g_pidhash[$_pid].tcb
+    set $_tcb = g_pidhash[$_pid]
     if ($_tcb->pid == $_pid)
         set $_regs = $_tcb->xcp.regs
         printf "PID %d NAME %s\n", $_pid, $_tcb.name
diff --git a/tools/nuttx-gdbinit b/tools/nuttx-gdbinit
index 2875fa05db..89bb04ed46 100644
--- a/tools/nuttx-gdbinit
+++ b/tools/nuttx-gdbinit
@@ -54,7 +54,7 @@ define _examine_target
   if ($_target_examined == 0x0)
     _examine_arch
 
-    set $_tcb0 = g_pidhash[0].tcb
+    set $_tcb0 = g_pidhash[0]
     set $_xcp_nregs = sizeof($_tcb0->xcp.regs) / sizeof($_tcb0->xcp.regs[0])
     set $_target_has_fpu = 0
 
@@ -385,7 +385,7 @@ define nxthread
   _examine_target
   _save_current_tcb
   set $hash = ($arg0 & ($_target_max_tasks - 1))
-  set $tcb = g_pidhash[$hash].tcb
+  set $tcb = g_pidhash[$hash]
   if ($tcb != 0x0)
     _print_thread $tcb
     if ($argc == 1)