arch/risc-v: Implement up_this_task using Thread Pointer (TP)

Summary:
- Added up_this_task() and up_update_task() macros to use TP register for fast task pointer access
- Modified CPU startup and initial state code to initialize TP with task pointer
- Updated context save/restore macros to handle TP only when TLS is enabled
- Added irq.h include to riscv_cpustart.c for new macros

Impact:
- Improves performance by using TP for fast task pointer access
- Reduces overhead of getting current task pointer in scheduler
- Maintains compatibility with TLS configuration
- No impact when CONFIG_SCHED_THREAD_LOCAL is enabled
- Changes context save/restore behavior for TP

Signed-off-by: Huang Qi <[email protected]>

arch/risc-v/include/irq.h

@@ -749,6 +749,33 @@ int up_this_cpu(void);
  * Inline Functions
  ****************************************************************************/
 
+/****************************************************************************
+ * Schedule acceleration macros
+ ****************************************************************************/
+
+/* When thread local storage is disabled (the usual case), the TP (Thread Pointer)
+ * register is available for use as a general purpose register. We use it to store
+ * the current task's TCB pointer for quick access.
+ *
+ * Note: If you want to use TLS (CONFIG_SCHED_THREAD_LOCAL), your toolchain must
+ * be compiled with --enable-tls option to properly support thread-local storage.
+ */
+
+#ifndef CONFIG_SCHED_THREAD_LOCAL
+#define up_this_task()                           \
+  ({                                             \
+    struct tcb_s* t;                             \
+    __asm__ __volatile__("mv %0, tp" : "=r"(t)); \
+    t;                                           \
+  })
+
+/* Update the current task pointer stored in TP register */
+#define up_update_task(t)                         \
+  {                                               \
+    __asm__ __volatile__("mv tp, %0" : : "r"(t)); \
+  }
+#endif /* CONFIG_SCHED_THREAD_LOCAL */
+
 /****************************************************************************
  * Name: up_irq_save
  *

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

@@ -36,6 +36,7 @@
 #include <nuttx/arch.h>
 #include <nuttx/spinlock.h>
 #include <nuttx/sched_note.h>
+#include <arch/irq.h>
 
 #include "sched/sched.h"
 #include "init/init.h"
@@ -69,6 +70,8 @@
 
 void riscv_cpu_boot(int cpu)
 {
+  struct tcb_s *tcb;
+
   /* Clear IPI for CPU(cpu) */
 
   riscv_ipi_clear(cpu);
@@ -100,9 +103,9 @@ void riscv_cpu_boot(int cpu)
 
   sinfo("CPU%d Started\n", this_cpu());
 
-#ifdef CONFIG_STACK_COLORATION
-  struct tcb_s *tcb = this_task();
+  tcb = current_task(this_cpu());
 
+#ifdef CONFIG_STACK_COLORATION
   /* If stack debug is enabled, then fill the stack with a
    * recognizable value that we can use later to test for high
    * water marks.
@@ -111,6 +114,8 @@ void riscv_cpu_boot(int cpu)
   riscv_stack_color(tcb->stack_alloc_ptr, 0);
 #endif
 
+  up_update_task(tcb);
+
   /* TODO: Setup FPU */
 
   /* Clear machine software interrupt for CPU(cpu) */

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

@@ -153,6 +153,8 @@ void up_initial_state(struct tcb_s *tcb)
 #ifdef CONFIG_SCHED_THREAD_LOCAL
   xcp->regs[REG_TP]      = (uintptr_t)tcb->stack_alloc_ptr +
                                      sizeof(struct tls_info_s);
+#else
+  xcp->regs[REG_TP]      = (uintptr_t)tcb;
 #endif
 
   /* Set the initial value of the interrupt context register.

arch/risc-v/src/common/riscv_macros.S

@@ -58,7 +58,9 @@
 #ifdef RISCV_SAVE_GP
   REGSTORE   x3,  REG_X3(\in)    /* gp */
 #endif
+#ifdef CONFIG_SCHED_THREAD_LOCAL
   REGSTORE   x4,  REG_X4(\in)    /* tp */
+#endif
   REGSTORE   x5,  REG_X5(\in)    /* t0 */
   REGSTORE   x6,  REG_X6(\in)    /* t1 */
   REGSTORE   x7,  REG_X7(\in)    /* t2 */
@@ -201,7 +203,9 @@
 #ifdef RISCV_SAVE_GP
   REGLOAD    x3,  REG_X3(\out)   /* gp */
 #endif
+#ifdef CONFIG_SCHED_THREAD_LOCAL
   REGLOAD    x4,  REG_X4(\out)   /* tp */
+#endif
   REGLOAD    x5,  REG_X5(\out)   /* t0 */
   REGLOAD    x6,  REG_X6(\out)   /* t1 */
   REGLOAD    x7,  REG_X7(\out)   /* t2 */