Move shadow space reservation to x86_64 makecontext

The shadow space (or home space) is a requirement of the x64 call
convention: we must reserve 32 bytes before the return address on the
stack. It only applies to x64, not to arm64 for example.

We don't have to deal with this when creating the stack. This is a
consideration for each individual `makecontext` to handle.

src/fiber.cr

@@ -108,21 +108,9 @@ class Fiber
 
     fiber_main = ->(f : Fiber) { f.run }
 
-    # FIXME: This line shouldn't be necessary (#7975)
-    stack_ptr = nil
-    {% if flag?(:win32) %}
-      # align stack bottom to 16 bytes
-      @stack_bottom = Pointer(Void).new(@stack_bottom.address & ~0x0f_u64)
-
-      # It's the caller's responsibility to allocate 32 bytes of "shadow space" on the stack right
-      # before calling the function (regardless of the actual number of parameters used)
-
-      stack_ptr = @stack_bottom - sizeof(Void*) * 6
-    {% else %}
-      # point to first addressable pointer on the stack (@stack_bottom points past
-      # the stack because the stack grows down):
-      stack_ptr = @stack_bottom - sizeof(Void*)
-    {% end %}
+    # point to first addressable pointer on the stack (@stack_bottom points past
+    # the stack because the stack grows down):
+    stack_ptr = @stack_bottom - sizeof(Void*)
 
     # align the stack pointer to 16 bytes:
     stack_ptr = Pointer(Void*).new(stack_ptr.address & ~0x0f_u64)

src/fiber/context/x86_64-microsoft.cr

@@ -6,14 +6,15 @@ class Fiber
     # A great explanation on stack contexts for win32:
     # https://web.archive.org/web/20220527113808/https://cfsamson.gitbook.io/green-threads-explained-in-200-lines-of-rust/supporting-windows
 
-    # 8 registers + 3 qwords for NT_TIB + 1 parameter + 10 128bit XMM registers
-    @context.stack_top = (stack_ptr - (12 + 10*2)).as(Void*)
+    # 4 shadow space + (8 registers + 3 qwords for NT_TIB + 1 parameter) + 10 128bit XMM registers
+    @context.stack_top = (stack_ptr - (4 + 12 + 10*2)).as(Void*)
     @context.resumable = 1
 
     # actual stack top, not including guard pages and reserved pages
     LibC.GetNativeSystemInfo(out system_info)
     stack_top = @stack_bottom - system_info.dwPageSize
 
+    stack_ptr -= 4                    # shadow space (or home space) before return address
     stack_ptr[0] = fiber_main.pointer # %rbx: Initial `resume` will `ret` to this address
     stack_ptr[-1] = self.as(Void*)    # %rcx: puts `self` as first argument for `fiber_main`