diff --git a/include/stddef.h b/include/stddef.h
index 3d328daa..6a56f6e0 100644
--- a/include/stddef.h
+++ b/include/stddef.h
@@ -19,9 +19,7 @@ typedef union { long long __ll; long double __ld; } max_align_t;
 #undef offsetof
 #define offsetof(type, field) __builtin_offsetof(type, field)
 
-#if defined __i386__ || defined __x86_64__
 void *alloca(size_t size);
-#endif
 
 #endif
 
diff --git a/include/tccdefs.h b/include/tccdefs.h
index 5c6535a0..d7596ac6 100644
--- a/include/tccdefs.h
+++ b/include/tccdefs.h
@@ -347,12 +347,8 @@
     __MAYBE_REDIR(void*, calloc, (__SIZE_TYPE__, __SIZE_TYPE__))
     __MAYBE_REDIR(void*, memalign, (__SIZE_TYPE__, __SIZE_TYPE__))
     __MAYBE_REDIR(void, free, (void*))
-#if defined __i386__ || defined __x86_64__
     __BOTH(void*, alloca, (__SIZE_TYPE__))
     void *alloca(__SIZE_TYPE__);
-#else
-    __BUILTIN(void*, alloca, (__SIZE_TYPE__))
-#endif
     __BUILTIN(void, abort, (void))
     __BOUND(void, longjmp, ())
 #if !defined _WIN32
diff --git a/lib/Makefile b/lib/Makefile
index 4d02c337..942add92 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -35,12 +35,12 @@ endif
 
 XFLAGS += -I$(TOP)
 
-I386_O = libtcc1.o alloca-bt.o $(COMMON_O)
-X86_64_O = libtcc1.o alloca-bt.o $(COMMON_O)
+I386_O = libtcc1.o $(COMMON_O)
+X86_64_O = libtcc1.o $(COMMON_O)
 ARM_O = libtcc1.o armeabi.o armflush.o $(COMMON_O)
 ARM64_O = lib-arm64.o $(COMMON_O)
 RISCV64_O = lib-arm64.o $(COMMON_O)
-COMMON_O = stdatomic.o atomic.o builtin.o alloca.o
+COMMON_O = stdatomic.o atomic.o builtin.o alloca.o alloca-bt.o
 WIN_O = crt1.o crt1w.o wincrt1.o wincrt1w.o dllcrt1.o dllmain.o
 LIN_O = dsohandle.o
 OSX_O =
diff --git a/lib/alloca-bt.S b/lib/alloca-bt.S
index 30c36e62..9092fdb8 100644
--- a/lib/alloca-bt.S
+++ b/lib/alloca-bt.S
@@ -92,5 +92,109 @@ p3:
     ret
 #endif
 
+/* ---------------------------------------------- */
+#elif defined __arm__
+
+.globl _(__bound_alloca)
+_(__bound_alloca):
+    mov r1, r0
+    add r0, r0, #1
+    rsb sp, r0, sp
+    bic sp, sp, #7
+    mov r0, sp
+    push { lr }
+    bl _(__bound_new_region)
+    pop { lr }
+    mov r0, sp
+    mov pc, lr
+
+/* ---------------------------------------------- */
+#elif defined __aarch64__ || defined __arm64__
+
+.globl _(__bound_alloca)
+_(__bound_alloca):
+#ifdef __TINYC__
+    .int 0xaa0003e1
+    .int 0x91004000
+    .int 0x927cec00
+#ifdef _WIN32
+    .int 0xb4000160
+    .int 0xd2820002
+    .int 0xeb02001f
+    .int 0x540000c3
+    .int 0xcb2263e3
+    .int 0xf940007f
+    .int 0xcb2263ff
+    .int 0xcb020000
+    .int 0x17fffffa
+    .int 0xb4000040
+#endif
+    .int 0xcb2063ff
+    .int 0x910003e0
+    .int 0xa9bf7bfd
+    .reloc ., R_AARCH64_CALL26,  _(__bound_new_region)
+    .int 0x94000000
+    .int 0xa8c17bfd
+    .int 0x910003e0
+    .int 0xd65f03c0
+#else
+    mov x1, x0
+    add x0, x0, #16     // Round up to 16-byte boundary
+    and x0, x0, #-16    // Ensure 16-byte alignment
+#ifdef _WIN32
+    cbz x0, p100        // If size is 0, skip to return
+    // Windows requires page-wise allocation with stack probing
+    mov x2, #4096       // Page size = 4096 bytes
+
+p101:
+    cmp x0, x2          // Compare remaining size with page size
+    b.lo    p102        // If less than page, jump to remainder
+
+    // Probe first, then allocate
+    sub x3, sp, x2      // Calculate guard page address (sp - 4096)
+    ldr xzr, [x3]       // Touch guard page FIRST
+    sub sp, sp, x2      // THEN allocate the page
+
+    sub x0, x0, x2      // Decrement remaining size
+    b   p101            // Continue loop
+
+p102:
+    // Allocate remaining bytes (less than one page)
+    cbz x0, p100        // If no remaining bytes, skip
+    sub sp, sp, x0      // Allocate remaining space
+#else
+    // Non-Windows: simple one-time allocation
+    sub sp, sp, x0      // Allocate space on stack
+#endif
+
+p100:
+    mov x0, sp          // Return allocated address
+    stp x29, x30, [sp, #-16]!
+    bl _(__bound_new_region)
+    ldp x29, x30, [sp], #16
+    mov x0, sp          // Return allocated address
+    ret                 // Return to caller
+#endif
+
+/* ---------------------------------------------- */
+#elif defined __riscv
+
+.globl _(__bound_alloca)
+_(__bound_alloca):
+    mv     a1, a0
+    sub    sp, sp, a0
+    addi   sp, sp, -16
+    andi   sp, sp, -16
+    add    a0, sp, zero
+    addi   sp,sp,-16
+    sd     s0,0(sp)
+    sd     ra,8(sp)
+    jal    _(__bound_new_region)
+    ld     s0,0(sp)
+    ld     ra,8(sp)
+    addi   sp,sp,16
+    add    a0, sp, zero
+    ret
+
 /* ---------------------------------------------- */
 #endif
diff --git a/lib/alloca.S b/lib/alloca.S
index f37be59d..e42abc1c 100644
--- a/lib/alloca.S
+++ b/lib/alloca.S
@@ -70,25 +70,18 @@ p3:
 /* ---------------------------------------------- */
 #elif defined __arm__
 
-    .text
-    .align  2
-    .global alloca
-    .type   alloca, %function
-alloca:
+.globl _(alloca)
+_(alloca):
     rsb sp, r0, sp
     bic sp, sp, #7
     mov r0, sp
     mov pc, lr
-    .size   alloca, .-alloca
 
 /* ---------------------------------------------- */
 #elif defined __aarch64__ || defined __arm64__
 
-    .text
-    .align  2
-    .global alloca
-    .type   alloca, %function
-alloca:
+.globl _(alloca)
+_(alloca):
 #ifdef __TINYC__
     .int 0x91003c00
     .int 0x927cec00
@@ -102,16 +95,16 @@ alloca:
     .int 0xcb2163ff
     .int 0xcb010000
     .int 0x17fffffa
-#endif
     .int 0xb4000040
+#endif
     .int 0xcb2063ff
     .int 0x910003e0
     .int 0xd65f03c0
 #else
     add x0, x0, #15     // Round up to 16-byte boundary
     and x0, x0, #-16    // Ensure 16-byte alignment
-    cbz x0, p100        // If size is 0, skip to return
 #ifdef _WIN32
+    cbz x0, p100        // If size is 0, skip to return
     // Windows requires page-wise allocation with stack probing
     mov x1, #4096       // Page size = 4096 bytes
 
@@ -140,20 +133,16 @@ p100:
     mov x0, sp          // Return allocated address
     ret                 // Return to caller
 #endif
-    .size   alloca, .-alloca
+
 /* ---------------------------------------------- */
 #elif defined __riscv
 
-	.text
-	.align	2
-	.global	alloca
-	.type	alloca, %function
-alloca:
-	sub	sp, sp, a0
-	addi	sp, sp, -15
-	andi	sp, sp, -16
-	add	a0, sp, zero
-	ret
-	.size	alloca, .-alloca
-/* ---------------------------------------------- */
+.globl _(alloca)
+_(alloca):
+    sub    sp, sp, a0
+    addi   sp, sp, -15
+    andi   sp, sp, -16
+    add    a0, sp, zero
+    ret
+
 #endif
diff --git a/libtcc.c b/libtcc.c
index 8809e617..01c4c7fd 100644
--- a/libtcc.c
+++ b/libtcc.c
@@ -40,7 +40,7 @@
 #elif defined(TCC_TARGET_ARM64)
 #include "arm64-gen.c"
 #include "arm64-link.c"
-#include "arm-asm.c"
+#include "arm64-asm.c"
 #elif defined(TCC_TARGET_C67)
 #include "c67-gen.c"
 #include "c67-link.c"
diff --git a/tcc.h b/tcc.h
index 310409df..058507fe 100644
--- a/tcc.h
+++ b/tcc.h
@@ -385,7 +385,7 @@ extern long double strtold (const char *__nptr, char **__endptr);
 #elif defined TCC_TARGET_ARM64
 # include "arm64-gen.c"
 # include "arm64-link.c"
-# include "arm-asm.c"
+# include "arm64-asm.c"
 #elif defined TCC_TARGET_C67
 # define TCC_TARGET_COFF
 # include "coff.h"
diff --git a/tccasm.c b/tccasm.c
index b6fc34e5..1c0e1b86 100644
--- a/tccasm.c
+++ b/tccasm.c
@@ -984,6 +984,23 @@ static void asm_parse_directive(TCCState *s1, int global)
         skip('@');
 	next();
 	break;
+    case TOK_ASMDIR_reloc:
+	{
+	    ExprValue e;
+
+	    next();
+	    asm_expr(s1, &e);
+	    skip(',');
+#ifdef __aarch64__
+	    if (strcmp(get_tok_str(tok, NULL), "R_AARCH64_CALL26"))
+#endif
+	        tcc_error("unimp: reloc '%s' unknown", get_tok_str(tok, NULL));
+	    next();
+	    skip(',');
+	    greloca(cur_text_section, get_asm_sym(tok, NULL), e.v, R_AARCH64_CALL26, 0);
+	    next();
+	}
+	break;
     default:
         tcc_error("unknown assembler directive '.%s'", get_tok_str(tok, NULL));
         break;
diff --git a/tccgen.c b/tccgen.c
index f8e81b81..0caeb151 100644
--- a/tccgen.c
+++ b/tccgen.c
@@ -1701,10 +1701,8 @@ ST_FUNC void gbound_args(int nb_args)
             gfunc_call(1);
             func_bound_add_epilog = 1;
         }
-#if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
         if (v == TOK_alloca)
             func_bound_add_epilog = 1;
-#endif
 #if TARGETOS_NetBSD
         if (v == TOK_longjmp) /* undo rename to __longjmp14 */
             sv->sym->asm_label = TOK___bound_longjmp;
diff --git a/tcctok.h b/tcctok.h
index 49dce0b2..4a16a22f 100644
--- a/tcctok.h
+++ b/tcctok.h
@@ -301,9 +301,7 @@
      DEF(TOK___fixxfdi, "__fixxfdi")
 #endif
 
-#if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
      DEF(TOK_alloca, "alloca")
-#endif
 
 #if defined TCC_TARGET_PE
      DEF(TOK___chkstk, "__chkstk")
@@ -415,6 +413,7 @@
  DEF_ASMDIR(long)
  DEF_ASMDIR(int)
  DEF_ASMDIR(symver)
+ DEF_ASMDIR(reloc)
  DEF_ASMDIR(section)    /* must be last directive */
 
 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
diff --git a/tests/boundtest.c b/tests/boundtest.c
index 8f150bbe..c92b93c2 100644
--- a/tests/boundtest.c
+++ b/tests/boundtest.c
@@ -169,14 +169,6 @@ int test13(void)
     return strlen(tab);
 }
 
-#if defined __i386__ || defined __x86_64__
-#define allocf(x)
-#else
-#undef alloca
-#define alloca(x) malloc(x)
-#define allocf(x) free(x)
-#endif
-
 int test14(void)
 {
     char *p = alloca(TAB_SIZE);
@@ -184,7 +176,6 @@ int test14(void)
     memset(p, 'a', TAB_SIZE);
     p[TAB_SIZE-1] = 0;
     ret = strlen(p);
-    allocf(p);
     return ret;
 }
 
@@ -196,7 +187,6 @@ int test15(void)
     memset(p, 'a', TAB_SIZE);
     p[TAB_SIZE-1] = 0;
     ret = strlen(p);
-    allocf(p);
     return ret;
 }
 
@@ -211,8 +201,6 @@ int test16()
 
     /* Test alloca embedded in a larger expression */
     printf("alloca : %s : %s\n", p, strcpy(q=alloca(strlen(demo)+1),demo) );
-    allocf(p);
-    allocf(q);
 
     return 0;
 }
@@ -228,8 +216,6 @@ int test17()
 
     /* Test alloca embedded in a larger expression */
     printf("alloca : %s : %s\n", p, strcpy(q=alloca(strlen(demo)),demo) );
-    allocf(p);
-    allocf(q);
 
     return 0;
 }
diff --git a/tests/tcctest.c b/tests/tcctest.c
index 6a16e945..3ddefaca 100644
--- a/tests/tcctest.c
+++ b/tests/tcctest.c
@@ -2941,14 +2941,12 @@ void old_style_function_test(void)
 
 void alloca_test()
 {
-#if defined __i386__ || defined __x86_64__ || defined __arm__
     char *p = alloca(16);
     strcpy(p,"123456789012345");
     printf("alloca: p is %s\n", p);
     char *demo = "This is only a test.\n";
     /* Test alloca embedded in a larger expression */
     printf("alloca: %s\n", strcpy(alloca(strlen(demo)+1),demo) );
-#endif
 }
 
 void *bounds_checking_is_enabled()
@@ -4208,7 +4206,6 @@ double get100 () { return 100.0; }
 
 void callsave_test(void)
 {
-#if defined __i386__ || defined __x86_64__ || defined __arm__
   int i, s; double *d; double t;
   s = sizeof (double);
   printf ("callsavetest: %d\n", s);
@@ -4221,7 +4218,6 @@ void callsave_test(void)
      generates a segfault.  */
   i = d[0] > get100 ();
   printf ("%d\n", i);
-#endif
 }