arm,riscv: eliminate idle_thread function prologue

Similar to #510 but for all other platforms. The idle_thread runs
without a stack and so cannot handle the stack prologue. This should
hopefully make the kernel rely less on FORCE_INLINE for this as well.

We create idle.S assembly files for each platform, as GCC does not
support `__attribute__((naked))` on AArch64 (GCC 13.2.0) and bails out.

Signed-off-by: julia <git.ts@trainwit.ch>

include/arch/arm/armv/armv7-a/armv/machine.h

@@ -8,12 +8,6 @@
 
 #include <util.h>
 
-/* See idle_thread for an explanation as to why FORCE_INLINE is required here. */
-static inline void FORCE_INLINE wfi(void)
-{
-    asm volatile("wfi" ::: "memory");
-}
-
 static inline void dsb(void)
 {
     asm volatile("dsb" ::: "memory");

include/arch/arm/armv/armv8-a/64/armv/machine.h

@@ -6,11 +6,6 @@
 
 #pragma once
 
-static inline void wfi(void)
-{
-    asm volatile("wfi" ::: "memory");
-}
-
 static inline void dsb(void)
 {
     asm volatile("dsb sy" ::: "memory");
@@ -42,4 +37,3 @@ static inline void isb(void)
 #define SYSTEM_READ_WORD(reg, v)  MRS(reg, v)
 #define SYSTEM_WRITE_64(reg, v)   MSR(reg, v)
 #define SYSTEM_READ_64(reg, v)    MRS(reg, v)
-

include/kernel/thread.h

@@ -165,7 +165,7 @@ void Arch_switchToIdleThread(void);
 void Arch_configureIdleThread(tcb_t *tcb);
 void Arch_activateIdleThread(tcb_t *tcb);
 
-void idle_thread(void);
+void NORETURN idle_thread(void);
 
 void configureIdleThread(tcb_t *tcb);
 void activateThread(void);
@@ -316,4 +316,3 @@ static inline void setThreadStateBlockedOnReply(tcb_t *tptr, reply_t *reply)
     scheduleTCB(tptr);
 }
 #endif
-

src/arch/arm/32/config.cmake

@@ -20,5 +20,5 @@ add_sources(
         idle.c
         kernel/thread.c
         kernel/vspace.c
-    ASMFILES head.S traps.S hyp_traps.S
+    ASMFILES head.S traps.S hyp_traps.S idle.S
 )

src/arch/arm/32/idle.S

@@ -0,0 +1,15 @@
+/*
+ * Copyright 2025, UNSW
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ */
+
+#include <machine/assembler.h>
+
+.code 32
+.section .text, "ax"
+
+BEGIN_FUNC(idle_thread)
+    1: wfi
+    b 1b
+END_FUNC(idle_thread)

src/arch/arm/32/idle.c

@@ -8,23 +8,6 @@
 #include <mode/machine.h>
 #include <api/debug.h>
 
-/*
- * The idle thread currently does not receive a stack pointer and so we rely on
- * optimisations for correctness here. More specifically, we assume:
- *  - Ordinary prologue/epilogue stack operations are optimised out
- *  - All nested function calls are inlined
- * Note that GCC does not correctly implement optimisation annotations on nested
- * functions, so FORCE_INLINE is required on the wfi declaration in this case.
- * Note that Clang doesn't obey FORCE_O2 and relies on the kernel being compiled
- * with optimisations enabled.
- */
-void FORCE_O2 idle_thread(void)
-{
-    while (1) {
-        wfi();
-    }
-}
-
 /** DONT_TRANSLATE */
 void NORETURN NO_INLINE VISIBLE halt(void)
 {

src/arch/arm/64/config.cmake

@@ -20,5 +20,5 @@ add_sources(
         idle.c
         kernel/thread.c
         kernel/vspace.c
-    ASMFILES head.S traps.S
+    ASMFILES head.S traps.S idle.S
 )

src/arch/arm/64/idle.S

@@ -0,0 +1,14 @@
+/*
+ * Copyright 2025, UNSW
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ */
+
+#include <machine/assembler.h>
+
+.section .text, "ax"
+
+BEGIN_FUNC(idle_thread)
+    1: wfi
+    b 1b
+END_FUNC(idle_thread)

src/arch/arm/64/idle.c

@@ -8,13 +8,6 @@
 #include <mode/machine.h>
 #include <api/debug.h>
 
-void idle_thread(void)
-{
-    while (1) {
-        wfi();
-    }
-}
-
 /** DONT_TRANSLATE */
 void NORETURN NO_INLINE VISIBLE halt(void)
 {

src/arch/riscv/config.cmake

@@ -122,7 +122,7 @@ add_sources(
         object/objecttype.c
         object/tcb.c
         smp/ipi.c
-    ASMFILES head.S traps.S
+    ASMFILES head.S traps.S idle.S
 )
 
 add_bf_source_old("KernelArchRiscV" "structures.bf" "include/arch/riscv" "arch/object")

src/arch/riscv/idle.S

@@ -0,0 +1,14 @@
+/*
+ * Copyright 2025, UNSW
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ */
+
+#include <machine/assembler.h>
+
+.section .text, "ax"
+
+BEGIN_FUNC(idle_thread)
+    1: wfi
+    j 1b
+END_FUNC(idle_thread)

src/arch/riscv/idle.c

@@ -8,13 +8,6 @@
 #include <config.h>
 #include <arch/sbi.h>
 
-void idle_thread(void)
-{
-    while (1) {
-        asm volatile("wfi");
-    }
-}
-
 /** DONT_TRANSLATE */
 void VISIBLE NO_INLINE halt(void)
 {

src/arch/x86/config.cmake

@@ -419,7 +419,7 @@ add_sources(
            machine/registerset.c
            benchmark/benchmark.c
            smp/ipi.c
-    ASMFILES multiboot.S
+    ASMFILES multiboot.S idle.S
 )
 
 add_bf_source_old("KernelArchX86" "structures.bf" "include/arch/x86" "arch/object")

src/arch/x86/idle.S

@@ -0,0 +1,15 @@
+/*
+ * Copyright 2025, UNSW
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ */
+
+#include <machine/assembler.h>
+
+.section .text, "ax"
+.code32
+
+BEGIN_FUNC(idle_thread)
+    1: hlt
+    jmp 1b
+END_FUNC(idle_thread)

src/arch/x86/idle.c

@@ -7,23 +7,6 @@
 #include <config.h>
 #include <api/debug.h>
 
-/*
- * The idle thread does not have a dedicated stack and runs in
- * the context of the idle thread TCB. Make sure that the compiler
- * always eliminates the function prologue by declaring the
- * idle_thread with the naked attribute.
- */
-__attribute__((naked)) NORETURN void idle_thread(void)
-{
-    /* We cannot use for-loop or while-loop here because they may
-     * involve stack manipulations (the compiler will not allow
-     * them in a naked function anyway). */
-    asm volatile(
-        "1: hlt\n"
-        "jmp 1b"
-    );
-}
-
 /** DONT_TRANSLATE */
 void VISIBLE halt(void)
 {