From 6fe1a6fe9f7086a77dd089ce1de67baa98044594 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?J=C5=8Dshin?= <git@wholezero.org>
Date: Thu, 21 Dec 2023 19:11:30 -0500
Subject: [PATCH] Reorder Launch arguments, pass aarch64 os

Third and fourth arguments are now identical between cosmo and Launch.
By passing sp as argument 4, we save a bit of register juggling.

Fourth argument (os) is now always passed by the loader on aarch64. It
is not yet processed by cosmo. Pushing this change separately, as the
cosmo side turns out to be somewhat more involved.
---
 ape/ape-m1.c   | 4 ++--
 ape/launch.S   | 9 ++++-----
 ape/loader.c   | 4 ++--
 libc/crt/crt.S | 1 +
 4 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/ape/ape-m1.c b/ape/ape-m1.c
index 874f452e1..4c32ebf85 100644
--- a/ape/ape-m1.c
+++ b/ape/ape-m1.c
@@ -726,10 +726,10 @@ __attribute__((__noreturn__)) static void Spawn(const char *exe, int fd,
 
   register long *x0 __asm__("x0") = sp;
   register char *x2 __asm__("x2") = path;
+  register int   x3 __asm__("x3") = 8;  /* _HOSTXNU */
   register struct Syslib *x15 __asm__("x15") = lib;
   register long x16 __asm__("x16") = e->e_entry;
   __asm__ volatile("mov\tx1,#0\n\t"
-                   "mov\tx3,#0\n\t"
                    "mov\tx4,#0\n\t"
                    "mov\tx5,#0\n\t"
                    "mov\tx6,#0\n\t"
@@ -758,7 +758,7 @@ __attribute__((__noreturn__)) static void Spawn(const char *exe, int fd,
                    "mov\tx0,#0\n\t"
                    "br\tx16"
                    : /* no outputs */
-                   : "r"(x0), "r"(x2), "r"(x15), "r"(x16)
+                   : "r"(x0), "r"(x2), "r"(x3), "r"(x15), "r"(x16)
                    : "memory");
   __builtin_unreachable();
 }
diff --git a/ape/launch.S b/ape/launch.S
index 87489f9f5..ae2cb58a1 100644
--- a/ape/launch.S
+++ b/ape/launch.S
@@ -32,15 +32,15 @@
 //	@param	rdi is passed through as-is
 //	@param	rsi is address of entrypoint (becomes zero)
 //	@param	rdx is passed through as-is
-//	@param	rcx is stack pointer (becomes r8)
+//	@param	rcx is passed through as-is
+//	@param	r8  is stack pointer (becomes zero)
 //	@noreturn
 Launch:
 #ifdef __aarch64__
 
 	mov	x16,x1
-	mov	sp,x3
+	mov	sp,x4
 	mov	x1,0
-	mov	x3,x4
 	mov	x4,0
 	mov	x5,0
 	mov	x6,0
@@ -70,8 +70,7 @@ Launch:
 
 #else
 
-	mov	%rcx,%rsp
-	mov	%r8,%rcx
+	mov	%r8,%rsp
 	xor	%r8d,%r8d
 	xor	%r9d,%r9d
 	xor	%r10d,%r10d
diff --git a/ape/loader.c b/ape/loader.c
index ece126229..ce9ef1b9c 100644
--- a/ape/loader.c
+++ b/ape/loader.c
@@ -225,7 +225,7 @@ struct ApeLoader {
 
 EXTERN_C long SystemCall(long, long, long, long, long, long, long, int);
 EXTERN_C void
-Launch(void *, long, void *, void *, int) __attribute__((__noreturn__));
+Launch(void *, long, void *, int, void *) __attribute__((__noreturn__));
 
 extern char __executable_start[];
 extern char _end[];
@@ -768,7 +768,7 @@ __attribute__((__noreturn__)) static void Spawn(int os, char *exe, int fd,
   Msyscall(dynbase + code, codesize, os);
 
   /* call program entrypoint */
-  Launch(IsFreebsd() ? sp : 0, dynbase + e->e_entry, exe, sp, os);
+  Launch(IsFreebsd() ? sp : 0, dynbase + e->e_entry, exe, os, sp);
 }
 
 static const char *TryElf(struct ApeLoader *M, union ElfEhdrBuf *ebuf,
diff --git a/libc/crt/crt.S b/libc/crt/crt.S
index 07978f47b..2f9fbb28c 100644
--- a/libc/crt/crt.S
+++ b/libc/crt/crt.S
@@ -143,6 +143,7 @@ _start:
 	mov	x1,x15
 
 //	third arg (x2) is the program path passed by ape-m1.c
+//	fourth arg (x3) is os
 
 //	switch to c code
 	bl	cosmo