Implement syscall(2)

include/muteki.h

@@ -17,6 +17,7 @@
 #include <muteki/ini.h>
 #include <muteki/loader.h>
 #include <muteki/memory.h>
+#include <muteki/syscall.h>
 #include <muteki/system.h>
 #include <muteki/threading.h>
 #include <muteki/ui.h>

include/muteki/syscall.h

@@ -0,0 +1,35 @@
+#ifndef __MUTEKI_SYSCALL_H__
+#define __MUTEKI_SYSCALL_H__
+
+#include <muteki/common.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define SYSCALL_HAS_STACK_ARGS (0x80000000ul)
+
+/**
+ * @brief Call arbitrary syscall with arbitrary number of arguments.
+ * @details
+ * This will reformat the stack frame, write an SVC instruction to an unused space of the stack, and jump to it.
+ *
+ * The higher 8 bits of the syscall number can encode special meanings. Currently setting the highest bit 
+ * (SYSCALL_HAS_STACK_ARGS) signals the function to take account of arguments allocated on the stack. That is,
+ * that bit should be set whenever there's more than 3 syscall arguments passed to this function.
+ *
+ * @warning Due to limitation of the Besta syscall scheme, this function is not thread-safe. Therefore to call it
+ * across multiple threads, a lock must be created and used across all threads that use this function.
+ *
+ * @param number The syscall number.
+ * @param ... Syscall arguments.
+ * @return Return value of the syscall. Can be casted to recover values of different type. Note that 64-bit values
+ * are not supported and only the lower 32-bit of such values can be retrieved.
+ */
+long syscall(long number, ...);
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+#endif // __MUTEKI_SYSCALL_H__

meson.build

@@ -46,7 +46,8 @@ syscalls_split = custom_target('syscalls_split',
 # Static library that bundles both sdk and krnl shims.
 static_library(
     'muteki-shims',
-    syscalls_split,
+    [syscalls_split, 'src/shims/syscall.c'],
+    include_directories: ['include/'],
     install : true,
     c_args : c_flags,
     link_args : ld_flags,

src/shims/syscall.c

@@ -0,0 +1,75 @@
+#include "muteki/syscall.h"
+
+#if defined(__clang__)
+
+// clang: Ignore unused parameter warning and clangd's warning on asm().
+#define NO_EXPECTED_WARNINGS_BEGIN \
+    _Pragma("clang diagnostic push") \
+    _Pragma("clang diagnostic ignored \"-Wunused-parameter\"") \
+    _Pragma("clang diagnostic ignored \"-Wlanguage-extension-token\"")
+#define NO_EXPECTED_WARNINGS_END \
+    _Pragma("clang diagnostic pop")
+
+#elif defined(__GNUC__)  // !defined(__clang__)
+
+// gcc: Ignore unused parameter warning.
+#define NO_EXPECTED_WARNINGS_BEGIN \
+    _Pragma("GCC diagnostic push") \
+    _Pragma("GCC diagnostic ignored \"-Wunused-parameter\"")
+#define NO_EXPECTED_WARNINGS_END \
+    _Pragma("GCC diagnostic pop")
+
+#else  // !defined(__GNUC__)
+
+// No-op
+#define NO_EXPECTED_WARNINGS_BEGIN
+#define NO_EXPECTED_WARNINGS_END
+
+#endif  // defined(__clang__) || defined(__GNUC__)
+
+__attribute__((used)) static unsigned int trampoline_area[] = {0xe1a00000};
+
+NO_EXPECTED_WARNINGS_BEGIN
+__attribute__((naked, target("arch=armv4t")))
+long syscall(long number, ...) {
+    // Starts as sysno a0 a1 a2 | a3 ... or sysno a0 a1 a2 | ...
+    asm(
+        // sysno x a1 a2 | a0 a3 ... or sysno x a1 a2 | a0 ...
+        "push {r1}\n\t"
+        // sysno a1 a2 x | a0 a3 ... or sysno a1 a2 x | a0 ...
+        "mov r1, r2\n\t"
+        "mov r2, r3\n\t"
+        // If passing 3+ arguments
+        "tst r0, #0x80000000\n\t"
+        "beq 1f\n\t"
+        // x a1 a2 x | sysno a0 a3 ...
+        "push {r0}\n\t"
+        // a0 a1 a2 x | sysno x a3 ...
+        "ldr r0, [sp, #4]\n\t"
+        // a0 a1 a2 a3 | sysno x x ...
+        "ldr r3, [sp, #8]\n\t"
+        // x a1 a2 a3 | sysno x a0 ...
+        "str r0, [sp, #8]\n\t"
+        // sysno a1 a2 a3 | x x a0 ...
+        "ldr r0, [sp]\n\t"
+        // sysno a1 a2 a3 | a0 ...
+        "add sp, sp, #8\n"
+        // Fall through
+
+        "1:\n\t"
+        // sysno x a2 a3? | a1 lr a0 ...
+        "push {r1, lr}\n\t"
+
+        "bic r0, r0, #0xff000000\n\t"
+        "orr r1, r0, #0xef000000\n\t"
+
+        // sysno svc a2 a3? | a1 lr a0 ...
+        "ldr r0, =trampoline_area\n\t"
+        // trampoline svc a2 a3? | a1 lr a0 ...
+        "str r1, [r0]\n\t"
+        // trampoline a1 a2 a3? | lr a0 ...
+        "pop {r1}\n\t"
+        "bx r0"
+    );
+}
+NO_EXPECTED_WARNINGS_END

src/syscall.s

@@ -0,0 +1,47 @@
+    .arm
+    .cpu arm7tdmi
+    .global syscall
+    .type syscall, %function
+
+@ Starts as sysno a0 a1 a2 | a3 ... or sysno a0 a1 a2 | ...
+syscall:
+    @ sysno x a1 a2 | a0 a3 ... or sysno x a1 a2 | a0 ...
+    push {r1}
+    @ sysno a1 a2 x | a0 a3 ... or sysno a1 a2 x | a0 ...
+    mov r1, r2
+    mov r2, r3
+    @ If passing 3+ arguments
+    tst r0, #0x80000000
+    beq _syscall_insert_lr
+    @ x a1 a2 x | sysno a0 a3 ...
+    push {r0}
+    @ a0 a1 a2 x | sysno x a3 ...
+    ldr r0, [sp, #4]
+    @ a0 a1 a2 a3 | sysno x x ...
+    ldr r3, [sp, #8]
+    @ x a1 a2 a3 | sysno x a0 ...
+    str r0, [sp, #8]
+    @ sysno a1 a2 a3 | x x a0 ...
+    ldr r0, [sp]
+    @ sysno a1 a2 a3 | a0 ...
+    add sp, sp, #8
+    @ Fall through
+
+_syscall_insert_lr:
+    @ sysno a1 a2 a3? | lr a0 ...
+    push {lr}
+
+    bic r0, r0, #0xff000000
+    orr r0, r0, #0xef000000
+
+    @ sysno a1 a2 a3? | trampoline[(svc sysno) nop nop] lr a0 ...
+    str r0, [sp, #-12]
+    ldr r0, =_nop
+    str r0, [sp, #-8]
+    str r0, [sp, #-4]
+    @ trampoline a1 a2 a3 | trampoline[(svc sysno) nop nop] lr a0 ...
+    sub r0, sp, #12
+    bx r0
+
+_nop:
+    mov r0, r0