From 1fad0f16541d1c9726cf9fc532078c31c384595b Mon Sep 17 00:00:00 2001
From: liwenxiang1 <liwenxiang1@xiaomi.com>
Date: Mon, 9 Dec 2024 14:09:45 +0800
Subject: [PATCH] arch/x86_64: this_task is stored in the CPU private data

By default in SMP, obtaining this_task requires disabling interrupts, obtaining the current CPU index, accessing a global variable, and re-enabling interrupts. Storing this_task in percpu makes retrieval faster.

Signed-off-by: liwenxiang1 <liwenxiang1@xiaomi.com>
---
 arch/x86_64/include/irq.h                     | 44 +++++++++----------
 .../x86_64/src/intel64/intel64_backtrace_fp.c |  4 +-
 arch/x86_64/src/intel64/intel64_cpustart.c    |  4 +-
 arch/x86_64/src/intel64/intel64_regdump.c     |  3 +-
 4 files changed, 28 insertions(+), 27 deletions(-)

diff --git a/arch/x86_64/include/irq.h b/arch/x86_64/include/irq.h
index 42790785bb528..0d82aafbd9339 100644
--- a/arch/x86_64/include/irq.h
+++ b/arch/x86_64/include/irq.h
@@ -72,13 +72,9 @@ struct intel64_cpu_s
 
   bool interrupt_context;
 
-/* current_regs holds a references to the current interrupt level
- * register storage structure.  If is non-NULL only during interrupt
- * processing.  Access to current_regs must be through
- * up_current_regs() and up_set_current_regs() functions
- */
+  /* Current task */
 
-  uint64_t *current_regs;
+  struct tcb_s *this_task;
 
 #ifdef CONFIG_LIB_SYSCALL
   /* Current user RSP for syscall */
@@ -143,22 +139,6 @@ static inline_function int up_cpu_index(void)
  * Inline functions
  ****************************************************************************/
 
-static inline_function uint64_t *up_current_regs(void)
-{
-  uint64_t *regs;
-  __asm__ volatile("movq %%gs:(%c1), %0"
-                   : "=r" (regs)
-                   : "i" (offsetof(struct intel64_cpu_s, current_regs)));
-  return regs;
-}
-
-static inline_function void up_set_current_regs(uint64_t *regs)
-{
-  __asm__ volatile("movq %0, %%gs:(%c1)"
-                   :: "r" (regs), "i" (offsetof(struct intel64_cpu_s,
-                                                current_regs)));
-}
-
 static inline_function bool up_interrupt_context(void)
 {
   bool flag;
@@ -177,12 +157,30 @@ static inline_function void up_set_interrupt_context(bool flag)
                                  interrupt_context)));
 }
 
+/****************************************************************************
+  * Schedule acceleration macros
+ ****************************************************************************/
+
+#define up_this_task()                                                  \
+  ({                                                                    \
+    struct tcb_s *this_task;                                            \
+    __asm__ volatile("movq %%gs:(%c1), %0"                              \
+                    : "=r" (this_task)                                  \
+                    : "i" (offsetof(struct intel64_cpu_s, this_task))); \
+    this_task;                                                          \
+  })
+
+#define up_update_task(t)                                               \
+  __asm__ volatile("movq %0, %%gs:(%c1)"                                \
+                   :: "r" ((struct tcb_s *)t),                          \
+                   "i" (offsetof(struct intel64_cpu_s, this_task)))
+
 /****************************************************************************
  * Name: up_getusrpc
  ****************************************************************************/
 
 #define up_getusrpc(regs) \
-    (((uint64_t *)((regs) ? (regs) : up_current_regs()))[REG_RIP])
+    (((uint64_t *)((regs) ? (regs) : running_regs()))[REG_RIP])
 
 #undef EXTERN
 #ifdef __cplusplus
diff --git a/arch/x86_64/src/intel64/intel64_backtrace_fp.c b/arch/x86_64/src/intel64/intel64_backtrace_fp.c
index 517a7b587e408..03dd357ffa1c4 100644
--- a/arch/x86_64/src/intel64/intel64_backtrace_fp.c
+++ b/arch/x86_64/src/intel64/intel64_backtrace_fp.c
@@ -145,8 +145,8 @@ int up_backtrace(struct tcb_s *tcb,
             {
               ret += backtrace(rtcb->stack_base_ptr,
                                rtcb->stack_base_ptr + rtcb->adj_stack_size,
-                               (void *)up_current_regs()[REG_RBP],
-                               (void *)up_current_regs()[REG_RIP],
+                               (void *)running_regs()[REG_RBP],
+                               (void *)running_regs()[REG_RIP],
                                &buffer[ret], size - ret, &skip);
             }
         }
diff --git a/arch/x86_64/src/intel64/intel64_cpustart.c b/arch/x86_64/src/intel64/intel64_cpustart.c
index 9671825d821fe..e2b46d4637557 100644
--- a/arch/x86_64/src/intel64/intel64_cpustart.c
+++ b/arch/x86_64/src/intel64/intel64_cpustart.c
@@ -148,7 +148,7 @@ void x86_64_ap_boot(void)
 
   x86_64_cpu_priv_set(cpu);
 
-  tcb = this_task();
+  tcb = current_task(cpu);
   UNUSED(tcb);
 
   /* Configure interrupts */
@@ -192,6 +192,8 @@ void x86_64_ap_boot(void)
       __revoke_low_memory();
     }
 
+  up_update_task(tcb);
+
   /* Then transfer control to the IDLE task */
 
   nx_idle_trampoline();
diff --git a/arch/x86_64/src/intel64/intel64_regdump.c b/arch/x86_64/src/intel64/intel64_regdump.c
index 9c13d57392b6a..646e1b32f86bc 100644
--- a/arch/x86_64/src/intel64/intel64_regdump.c
+++ b/arch/x86_64/src/intel64/intel64_regdump.c
@@ -30,6 +30,7 @@
 #include <debug.h>
 #include <nuttx/irq.h>
 
+#include "sched/sched.h"
 #include "x86_64_internal.h"
 
 /****************************************************************************
@@ -113,7 +114,7 @@ void backtrace(uint64_t rbp)
 
 void up_dump_register(void *dumpregs)
 {
-  volatile uint64_t *regs = dumpregs ? dumpregs : up_current_regs();
+  volatile uint64_t *regs = dumpregs ? dumpregs : running_regs();
   uint64_t mxcsr;
   uint64_t cr2;