[arm64] add hw thread self (#8942)

* [libcpu] arm64: Add hardware thread_self support

This patch introduces hardware-based thread self-identification
for the AArch64 architecture. It optimizes thread management by
using hardware registers to store and access the current thread's
pointer, reducing overhead and improving overall performance.

Changes include:
- Added `ARCH_USING_HW_THREAD_SELF` configuration option.
- Modified `rtdef.h`, `rtsched.h` to conditionally include
  `critical_switch_flag` based on the new config.
- Updated context management in `context_gcc.S`, `cpuport.h`
  to support hardware-based thread self.
- Enhanced `scheduler_mp.c` and `thread.c` to leverage the new
  hardware thread self feature.

These modifications ensure better scheduling and thread handling,
particularly in multi-core environments, by minimizing the
software overhead associated with thread management.

Signed-off-by: Shell <[email protected]>

* fixup: address suggestion

* fixup: rt_current_thread as global

* scheduler: add cpu object for UP scheduler

Also, maintain the rt_current_thread in cpu object on UP scheduler.

---------

Signed-off-by: Shell <[email protected]>

4 files changed, 111 insertions, 40 deletions

diff --git a/src/cpu_up.c b/src/cpu_up.c
index 8a1804cf55..32825461b0 100644
--- a/src/cpu_up.c
+++ b/src/cpu_up.c
@@ -6,10 +6,14 @@
  * Change Logs:
  * Date           Author       Notes
  * 2024-04-19     Shell        Fixup UP irq spinlock
+ * 2024-05-22     Shell        Add UP cpu object and
+ *                             maintain the rt_current_thread inside it
  */
 #include <rthw.h>
 #include <rtthread.h>
 
+static struct rt_cpu _cpu;
+
 /**
  * @brief   Initialize a static spinlock object.
  *
@@ -80,3 +84,25 @@ void rt_spin_unlock_irqrestore(struct rt_spinlock *lock, rt_base_t level)
     rt_exit_critical_safe(critical_level);
     rt_hw_interrupt_enable(level);
 }
+
+/**
+ * @brief   This fucntion will return current cpu object.
+ *
+ * @return  Return a pointer to the current cpu object.
+ */
+struct rt_cpu *rt_cpu_self(void)
+{
+    return &_cpu;
+}
+
+/**
+ * @brief   This fucntion will return the cpu object corresponding to index.
+ *
+ * @param   index is the index of target cpu object.
+ *
+ * @return  Return a pointer to the cpu object corresponding to index.
+ */
+struct rt_cpu *rt_cpu_index(int index)
+{
+    return index == 0 ? &_cpu : RT_NULL;
+}
diff --git a/src/scheduler_mp.c b/src/scheduler_mp.c
index c1626c5f99..7a7bd61e00 100644
--- a/src/scheduler_mp.c
+++ b/src/scheduler_mp.c
@@ -33,6 +33,7 @@
  * 2023-12-10     xqyjlj       use rt_hw_spinlock
  * 2024-01-05     Shell        Fixup of data racing in rt_critical_level
  * 2024-01-18     Shell        support rt_sched_thread of scheduling status for better mt protection
+ * 2024-01-18     Shell        support rt_hw_thread_self to improve overall performance
  */
 
 #include <rtthread.h>
@@ -99,6 +100,18 @@ static struct rt_spinlock _mp_scheduler_lock;
         rt_hw_local_irq_enable(level);     \
     } while (0)
 
+#ifdef ARCH_USING_HW_THREAD_SELF
+#define IS_CRITICAL_SWITCH_PEND(pcpu, curthr)  (RT_SCHED_CTX(curthr).critical_switch_flag)
+#define SET_CRITICAL_SWITCH_FLAG(pcpu, curthr) (RT_SCHED_CTX(curthr).critical_switch_flag = 1)
+#define CLR_CRITICAL_SWITCH_FLAG(pcpu, curthr) (RT_SCHED_CTX(curthr).critical_switch_flag = 0)
+
+#else /* !ARCH_USING_HW_THREAD_SELF */
+#define IS_CRITICAL_SWITCH_PEND(pcpu, curthr)  ((pcpu)->critical_switch_flag)
+#define SET_CRITICAL_SWITCH_FLAG(pcpu, curthr) ((pcpu)->critical_switch_flag = 1)
+#define CLR_CRITICAL_SWITCH_FLAG(pcpu, curthr) ((pcpu)->critical_switch_flag = 0)
+
+#endif /* ARCH_USING_HW_THREAD_SELF */
+
 static rt_uint32_t rt_thread_ready_priority_group;
 #if RT_THREAD_PRIORITY_MAX > 32
 /* Maximum priority level, 256 */
@@ -749,7 +762,7 @@ rt_err_t rt_sched_unlock_n_resched(rt_sched_lock_level_t level)
         /* leaving critical region of global context since we can't schedule */
         SCHEDULER_CONTEXT_UNLOCK(pcpu);
 
-        pcpu->critical_switch_flag = 1;
+        SET_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
         error = -RT_ESCHEDLOCKED;
 
         SCHEDULER_EXIT_CRITICAL(current_thread);
@@ -757,7 +770,7 @@ rt_err_t rt_sched_unlock_n_resched(rt_sched_lock_level_t level)
     else
     {
         /* flush critical switch flag since a scheduling is done */
-        pcpu->critical_switch_flag = 0;
+        CLR_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
 
         /* pick the highest runnable thread, and pass the control to it */
         to_thread = _prepare_context_switch_locked(cpu_id, pcpu, current_thread);
@@ -828,7 +841,7 @@ void rt_schedule(void)
     /* whether caller had locked the local scheduler already */
     if (RT_SCHED_CTX(current_thread).critical_lock_nest > 1)
     {
-        pcpu->critical_switch_flag = 1;
+        SET_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
 
         SCHEDULER_EXIT_CRITICAL(current_thread);
 
@@ -837,7 +850,7 @@ void rt_schedule(void)
     else
     {
         /* flush critical switch flag since a scheduling is done */
-        pcpu->critical_switch_flag = 0;
+        CLR_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
         pcpu->irq_switch_flag = 0;
 
         /**
@@ -912,13 +925,13 @@ void rt_scheduler_do_irq_switch(void *context)
     /* whether caller had locked the local scheduler already */
     if (RT_SCHED_CTX(current_thread).critical_lock_nest > 1)
     {
-        pcpu->critical_switch_flag = 1;
+        SET_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
         SCHEDULER_EXIT_CRITICAL(current_thread);
     }
     else if (rt_atomic_load(&(pcpu->irq_nest)) == 0)
     {
         /* flush critical & irq switch flag since a scheduling is done */
-        pcpu->critical_switch_flag = 0;
+        CLR_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
         pcpu->irq_switch_flag = 0;
 
         SCHEDULER_CONTEXT_LOCK(pcpu);
@@ -1056,6 +1069,9 @@ void rt_sched_post_ctx_switch(struct rt_thread *thread)
     }
     /* safe to access since irq is masked out */
     pcpu->current_thread = thread;
+#ifdef ARCH_USING_HW_THREAD_SELF
+    rt_hw_thread_set_self(thread);
+#endif /* ARCH_USING_HW_THREAD_SELF */
 }
 
 #ifdef RT_DEBUGING_CRITICAL
@@ -1096,14 +1112,28 @@ void rt_exit_critical_safe(rt_base_t critical_level)
 #endif /* RT_DEBUGING_CRITICAL */
 RTM_EXPORT(rt_exit_critical_safe);
 
+#ifdef ARCH_USING_HW_THREAD_SELF
+#define FREE_THREAD_SELF(lvl)
+
+#else /* !ARCH_USING_HW_THREAD_SELF */
+#define FREE_THREAD_SELF(lvl)        \
+    do                               \
+    {                                \
+        rt_hw_local_irq_enable(lvl); \
+    } while (0)
+
+#endif /* ARCH_USING_HW_THREAD_SELF */
+
 /**
  * @brief This function will lock the thread scheduler.
  */
 rt_base_t rt_enter_critical(void)
 {
-    rt_base_t level;
     rt_base_t critical_level;
     struct rt_thread *current_thread;
+
+#ifndef ARCH_USING_HW_THREAD_SELF
+    rt_base_t level;
     struct rt_cpu *pcpu;
 
     /* disable interrupt */
@@ -1111,9 +1141,15 @@ rt_base_t rt_enter_critical(void)
 
     pcpu = rt_cpu_self();
     current_thread = pcpu->current_thread;
+
+#else /* !ARCH_USING_HW_THREAD_SELF */
+    current_thread = rt_hw_thread_self();
+
+#endif /* ARCH_USING_HW_THREAD_SELF */
+
     if (!current_thread)
     {
-        rt_hw_local_irq_enable(level);
+        FREE_THREAD_SELF(level);
         /* scheduler unavailable */
         return -RT_EINVAL;
     }
@@ -1122,8 +1158,7 @@ rt_base_t rt_enter_critical(void)
     RT_SCHED_CTX(current_thread).critical_lock_nest++;
     critical_level = RT_SCHED_CTX(current_thread).critical_lock_nest;
 
-    /* enable interrupt */
-    rt_hw_local_irq_enable(level);
+    FREE_THREAD_SELF(level);
 
     return critical_level;
 }
@@ -1134,9 +1169,11 @@ RTM_EXPORT(rt_enter_critical);
  */
 void rt_exit_critical(void)
 {
-    rt_base_t level;
     struct rt_thread *current_thread;
     rt_bool_t need_resched;
+
+#ifndef ARCH_USING_HW_THREAD_SELF
+    rt_base_t level;
     struct rt_cpu *pcpu;
 
     /* disable interrupt */
@@ -1144,9 +1181,15 @@ void rt_exit_critical(void)
 
     pcpu = rt_cpu_self();
     current_thread = pcpu->current_thread;
+
+#else /* !ARCH_USING_HW_THREAD_SELF */
+    current_thread = rt_hw_thread_self();
+
+#endif /* ARCH_USING_HW_THREAD_SELF */
+
     if (!current_thread)
     {
-        rt_hw_local_irq_enable(level);
+        FREE_THREAD_SELF(level);
         return;
     }
 
@@ -1157,11 +1200,10 @@ void rt_exit_critical(void)
     if (RT_SCHED_CTX(current_thread).critical_lock_nest == 0)
     {
         /* is there any scheduling request unfinished? */
-        need_resched = pcpu->critical_switch_flag;
-        pcpu->critical_switch_flag = 0;
+        need_resched = IS_CRITICAL_SWITCH_PEND(pcpu, current_thread);
+        CLR_CRITICAL_SWITCH_FLAG(pcpu, current_thread);
 
-        /* enable interrupt */
-        rt_hw_local_irq_enable(level);
+        FREE_THREAD_SELF(level);
 
         if (need_resched)
             rt_schedule();
@@ -1171,8 +1213,7 @@ void rt_exit_critical(void)
         /* each exit_critical is strictly corresponding to an enter_critical */
         RT_ASSERT(RT_SCHED_CTX(current_thread).critical_lock_nest > 0);
 
-        /* enable interrupt */
-        rt_hw_local_irq_enable(level);
+        FREE_THREAD_SELF(level);
     }
 }
 RTM_EXPORT(rt_exit_critical);
diff --git a/src/scheduler_up.c b/src/scheduler_up.c
index 822d4a8046..acc8ee8c95 100644
--- a/src/scheduler_up.c
+++ b/src/scheduler_up.c
@@ -48,7 +48,6 @@ rt_uint8_t rt_thread_ready_table[32];
 
 extern volatile rt_uint8_t rt_interrupt_nest;
 static rt_int16_t rt_scheduler_lock_nest;
-struct rt_thread *rt_current_thread = RT_NULL;
 rt_uint8_t rt_current_priority;
 
 #if defined(RT_USING_HOOK) && defined(RT_HOOK_USING_FUNC_PTR)
@@ -175,7 +174,7 @@ void rt_system_scheduler_start(void)
 
     to_thread = _scheduler_get_highest_priority_thread(&highest_ready_priority);
 
-    rt_current_thread = to_thread;
+    rt_cpu_self()->current_thread = to_thread;
 
     rt_sched_remove_thread(to_thread);
     RT_SCHED_CTX(to_thread).stat = RT_THREAD_RUNNING;
@@ -203,6 +202,8 @@ void rt_schedule(void)
     rt_base_t level;
     struct rt_thread *to_thread;
     struct rt_thread *from_thread;
+    /* using local variable to avoid unecessary function call */
+    struct rt_thread *curr_thread = rt_thread_self();
 
     /* disable interrupt */
     level = rt_hw_interrupt_disable();
@@ -219,15 +220,16 @@ void rt_schedule(void)
 
             to_thread = _scheduler_get_highest_priority_thread(&highest_ready_priority);
 
-            if ((RT_SCHED_CTX(rt_current_thread).stat & RT_THREAD_STAT_MASK) == RT_THREAD_RUNNING)
+            if ((RT_SCHED_CTX(curr_thread).stat & RT_THREAD_STAT_MASK) == RT_THREAD_RUNNING)
             {
-                if (RT_SCHED_PRIV(rt_current_thread).current_priority < highest_ready_priority)
+                if (RT_SCHED_PRIV(curr_thread).current_priority < highest_ready_priority)
                 {
-                    to_thread = rt_current_thread;
+                    to_thread = curr_thread;
                 }
-                else if (RT_SCHED_PRIV(rt_current_thread).current_priority == highest_ready_priority && (RT_SCHED_CTX(rt_current_thread).stat & RT_THREAD_STAT_YIELD_MASK) == 0)
+                else if (RT_SCHED_PRIV(curr_thread).current_priority == highest_ready_priority
+                         && (RT_SCHED_CTX(curr_thread).stat & RT_THREAD_STAT_YIELD_MASK) == 0)
                 {
-                    to_thread = rt_current_thread;
+                    to_thread = curr_thread;
                 }
                 else
                 {
@@ -235,12 +237,12 @@ void rt_schedule(void)
                 }
             }
 
-            if (to_thread != rt_current_thread)
+            if (to_thread != curr_thread)
             {
                 /* if the destination thread is not the same as current thread */
                 rt_current_priority = (rt_uint8_t)highest_ready_priority;
-                from_thread         = rt_current_thread;
-                rt_current_thread   = to_thread;
+                from_thread                   = curr_thread;
+                rt_cpu_self()->current_thread = to_thread;
 
                 RT_OBJECT_HOOK_CALL(rt_scheduler_hook, (from_thread, to_thread));
 
@@ -282,11 +284,11 @@ void rt_schedule(void)
 #ifdef RT_USING_SIGNALS
                     /* check stat of thread for signal */
                     level = rt_hw_interrupt_disable();
-                    if (RT_SCHED_CTX(rt_current_thread).stat & RT_THREAD_STAT_SIGNAL_PENDING)
+                    if (RT_SCHED_CTX(curr_thread).stat & RT_THREAD_STAT_SIGNAL_PENDING)
                     {
                         extern void rt_thread_handle_sig(rt_bool_t clean_state);
 
-                        RT_SCHED_CTX(rt_current_thread).stat &= ~RT_THREAD_STAT_SIGNAL_PENDING;
+                        RT_SCHED_CTX(curr_thread).stat &= ~RT_THREAD_STAT_SIGNAL_PENDING;
 
                         rt_hw_interrupt_enable(level);
 
@@ -310,8 +312,8 @@ void rt_schedule(void)
             }
             else
             {
-                rt_sched_remove_thread(rt_current_thread);
-                RT_SCHED_CTX(rt_current_thread).stat = RT_THREAD_RUNNING | (RT_SCHED_CTX(rt_current_thread).stat & ~RT_THREAD_STAT_MASK);
+                rt_sched_remove_thread(curr_thread);
+                RT_SCHED_CTX(curr_thread).stat = RT_THREAD_RUNNING | (RT_SCHED_CTX(curr_thread).stat & ~RT_THREAD_STAT_MASK);
             }
         }
     }
diff --git a/src/thread.c b/src/thread.c
index 8d4752409d..1369b7560a 100644
--- a/src/thread.c
+++ b/src/thread.c
@@ -355,20 +355,22 @@ RTM_EXPORT(rt_thread_init);
  */
 rt_thread_t rt_thread_self(void)
 {
-#ifdef RT_USING_SMP
-    rt_base_t lock;
+#ifndef RT_USING_SMP
+    return rt_cpu_self()->current_thread;
+
+#elif defined (ARCH_USING_HW_THREAD_SELF)
+    return rt_hw_thread_self();
+
+#else /* !ARCH_USING_HW_THREAD_SELF */
     rt_thread_t self;
+    rt_base_t lock;
 
     lock = rt_hw_local_irq_disable();
     self = rt_cpu_self()->current_thread;
     rt_hw_local_irq_enable(lock);
-    return self;
-
-#else /* !RT_USING_SMP */
-    extern rt_thread_t rt_current_thread;
 
-    return rt_current_thread;
-#endif /* RT_USING_SMP */
+    return self;
+#endif /* ARCH_USING_HW_THREAD_SELF */
 }
 RTM_EXPORT(rt_thread_self);