kernel: arch: move arch_swap() declaration

Moves the arch_swap() declaration out of kernel_arch_interface.h
and into the various architectures' kernel_arch_func.h. This
permits the arch_swap() to be inlined on ARM, but extern'd on
the other architectures that still implement arch_swap().

Inlining this function on ARM has shown at least a +5% performance
boost according to the thread_metric benchmark on the disco_l475_iot1
board.

Signed-off-by: Peter Mitsis <peter.mitsis@intel.com>

arch/arm/core/cortex_a_r/CMakeLists.txt

@@ -24,4 +24,4 @@ zephyr_library_sources_ifdef(CONFIG_SEMIHOST semihost.c)
 zephyr_library_sources_ifdef(CONFIG_THREAD_LOCAL_STORAGE __aeabi_read_tp.S)
 zephyr_library_sources_ifdef(CONFIG_ARCH_CACHE cache.c)
 zephyr_library_sources_ifdef(CONFIG_USE_SWITCH switch.S)
-zephyr_library_sources_ifndef(CONFIG_USE_SWITCH swap.c swap_helper.S exc_exit.S)
+zephyr_library_sources_ifndef(CONFIG_USE_SWITCH swap_helper.S exc_exit.S)

arch/arm/core/cortex_a_r/swap.c

@@ -1,30 +0,0 @@
-/*
- * Copyright (c) 2018 Linaro, Limited
- *
- * SPDX-License-Identifier: Apache-2.0
- */
-
-#include <zephyr/kernel.h>
-#include <kernel_internal.h>
-
-#include <errno.h>
-
-/* The 'key' actually represents the BASEPRI register
- * prior to disabling interrupts via the BASEPRI mechanism.
- *
- * arch_swap() itself does not do much.
- */
-int arch_swap(unsigned int key)
-{
-	/* store off key and return value */
-	arch_current_thread()->arch.basepri = key;
-	arch_current_thread()->arch.swap_return_value = -EAGAIN;
-
-	z_arm_cortex_r_svc();
-	irq_unlock(key);
-
-	/* Context switch is performed here. Returning implies the
-	 * thread has been context-switched-in again.
-	 */
-	return arch_current_thread()->arch.swap_return_value;
-}

arch/arm/core/cortex_m/CMakeLists.txt

@@ -11,7 +11,6 @@ zephyr_library_sources(
   scb.c
   thread_abort.c
   vector_table.S
-  swap.c
   swap_helper.S
   irq_manage.c
   prep_c.c

arch/arm/core/cortex_m/swap.c

@@ -1,49 +0,0 @@
-/*
- * Copyright (c) 2018 Linaro, Limited
- *
- * SPDX-License-Identifier: Apache-2.0
- */
-
-#include <zephyr/kernel.h>
-#include <kernel_internal.h>
-
-#include <errno.h>
-
-/* The 'key' actually represents the BASEPRI register
- * prior to disabling interrupts via the BASEPRI mechanism.
- *
- * arch_swap() itself does not do much.
- *
- * It simply stores the intlock key (the BASEPRI value) parameter into
- * current->basepri, and then triggers a PendSV exception, which does
- * the heavy lifting of context switching.
-
- * This is the only place we have to save BASEPRI since the other paths to
- * z_arm_pendsv all come from handling an interrupt, which means we know the
- * interrupts were not locked: in that case the BASEPRI value is 0.
- *
- * Given that arch_swap() is called to effect a cooperative context switch,
- * only the caller-saved integer registers need to be saved in the thread of the
- * outgoing thread. This is all performed by the hardware, which stores it in
- * its exception stack frame, created when handling the z_arm_pendsv exception.
- *
- * On ARMv6-M, the intlock key is represented by the PRIMASK register,
- * as BASEPRI is not available.
- */
-int arch_swap(unsigned int key)
-{
-	/* store off key and return value */
-	arch_current_thread()->arch.basepri = key;
-	arch_current_thread()->arch.swap_return_value = -EAGAIN;
-
-	/* set pending bit to make sure we will take a PendSV exception */
-	SCB->ICSR |= SCB_ICSR_PENDSVSET_Msk;
-
-	/* clear mask or enable all irqs to take a pendsv */
-	irq_unlock(0);
-
-	/* Context switch is performed here. Returning implies the
-	 * thread has been context-switched-in again.
-	 */
-	return arch_current_thread()->arch.swap_return_value;
-}

arch/arm/include/cortex_a_r/kernel_arch_func.h

@@ -37,6 +37,21 @@ static ALWAYS_INLINE void arch_kernel_init(void)
 
 #ifndef CONFIG_USE_SWITCH
 
+static ALWAYS_INLINE int arch_swap(unsigned int key)
+{
+	/* store off key and return value */
+	arch_current_thread()->arch.basepri = key;
+	arch_current_thread()->arch.swap_return_value = -EAGAIN;
+
+	z_arm_cortex_r_svc();
+	irq_unlock(key);
+
+	/* Context switch is performed here. Returning implies the
+	 * thread has been context-switched-in again.
+	 */
+	return arch_current_thread()->arch.swap_return_value;
+}
+
 static ALWAYS_INLINE void
 arch_thread_return_value_set(struct k_thread *thread, unsigned int value)
 {

arch/arm/include/cortex_m/kernel_arch_func.h

@@ -84,6 +84,25 @@ extern FUNC_NORETURN void z_arm_userspace_enter(k_thread_entry_t user_entry,
 
 extern void z_arm_fatal_error(unsigned int reason, const struct arch_esf *esf);
 
+static ALWAYS_INLINE int arch_swap(unsigned int key)
+{
+	/* store off key and return value */
+	arch_current_thread()->arch.basepri = key;
+	arch_current_thread()->arch.swap_return_value = -EAGAIN;
+
+	/* set pending bit to make sure we will take a PendSV exception */
+	SCB->ICSR |= SCB_ICSR_PENDSVSET_Msk;
+
+	/* clear mask or enable all irqs to take a pendsv */
+	irq_unlock(0);
+
+	/* Context switch is performed here. Returning implies the
+	 * thread has been context-switched-in again.
+	 */
+	return arch_current_thread()->arch.swap_return_value;
+}
+
+
 #endif /* _ASMLANGUAGE */
 
 #ifdef __cplusplus

arch/mips/include/kernel_arch_func.h

@@ -47,6 +47,8 @@ static inline bool arch_is_in_isr(void)
 	return _current_cpu->nested != 0U;
 }
 
+int arch_swap(unsigned int key);
+
 #ifdef CONFIG_IRQ_OFFLOAD
 void z_irq_do_offload(void);
 #endif

arch/nios2/include/kernel_arch_func.h

@@ -51,6 +51,8 @@ static inline bool arch_is_in_isr(void)
 	return _kernel.cpus[0].nested != 0U;
 }
 
+int arch_swap(unsigned int key);
+
 #ifdef CONFIG_IRQ_OFFLOAD
 void z_irq_do_offload(void);
 #endif

arch/posix/include/kernel_arch_func.h

@@ -42,6 +42,8 @@ static inline bool arch_is_in_isr(void)
 	return _kernel.cpus[0].nested != 0U;
 }
 
+int arch_swap(unsigned int key);
+
 #endif /* _ASMLANGUAGE */
 
 #endif /* ZEPHYR_ARCH_POSIX_INCLUDE_KERNEL_ARCH_FUNC_H_ */

arch/x86/include/ia32/kernel_arch_func.h

@@ -37,6 +37,8 @@ arch_thread_return_value_set(struct k_thread *thread, unsigned int value)
 
 extern void arch_cpu_atomic_idle(unsigned int key);
 
+int arch_swap(unsigned int key);
+
 /* ASM code to fiddle with registers to enable the MMU with PAE paging */
 void z_x86_enable_paging(void);
 

kernel/include/kernel_arch_interface.h

@@ -129,7 +129,10 @@ void arch_new_thread(struct k_thread *thread, k_thread_stack_t *stack,
  *        location, which must be updated.
  */
 static inline void arch_switch(void *switch_to, void **switched_from);
-#else
+#endif /* CONFIG_USE_SWITCH */
+
+#if !defined(CONFIG_USE_SWITCH) || defined(__DOXYGEN__)
+#if defined(__DOXYGEN__)
 /**
  * Cooperatively context switch
  *
@@ -143,6 +146,7 @@ static inline void arch_switch(void *switch_to, void **switched_from);
  *         blocking operation.
  */
 int arch_swap(unsigned int key);
+#endif /* __DOXYGEN__ */
 
 /**
  * Set the return value for the specified thread.
@@ -154,7 +158,7 @@ int arch_swap(unsigned int key);
  */
 static ALWAYS_INLINE void
 arch_thread_return_value_set(struct k_thread *thread, unsigned int value);
-#endif /* CONFIG_USE_SWITCH */
+#endif /* !CONFIG_USE_SWITCH || __DOXYGEN__ */
 
 #ifdef CONFIG_ARCH_HAS_CUSTOM_SWAP_TO_MAIN
 /**

tests/arch/arm/arm_thread_swap/src/arm_thread_arch.c

@@ -429,6 +429,13 @@ static void alt_thread_entry(void *p1, void *p2, void *p3)
 		"Alternative thread: switch flag not false on thread exit\n");
 }
 
+#if !defined(CONFIG_NO_OPTIMIZATIONS)
+static int __noinline arch_swap_wrapper(void)
+{
+	return arch_swap(BASEPRI_MODIFIED_1);
+}
+#endif
+
 ZTEST(arm_thread_swap, test_arm_thread_swap)
 {
 	int test_flag;
@@ -609,9 +616,11 @@ ZTEST(arm_thread_swap, test_arm_thread_swap)
 
 	/* Fake a different irq_unlock key when performing swap.
 	 * This will be verified by the alternative test thread.
+	 *
+	 * Force an indirect call to arch_swap() to prevent the compiler from
+	 * changing the saved callee registers as arch_swap() is inlined.
 	 */
-	register int swap_return_val __asm__("r0") =
-		arch_swap(BASEPRI_MODIFIED_1);
+	register int swap_return_val __asm__("r0") = arch_swap_wrapper();
 
 #endif /* CONFIG_NO_OPTIMIZATIONS */