Feature: self hosting prototype implemented

This is a large change. In order to remove myself from libc's arcane
interface, I implemented an independent runtime layer. It is based on
musl's wonderful implementation mostly. Critically, if libc is linked to
the program, then we cooperate. Namely, we call start main and let libc
do all initialization. If not, then we have a noop defined in rt3.a.

The general structure of the file is:
1. sys/$os/$arch contains all architecture dependent code
2. sys/$os/port contains all code that depends on the os, but is
   portable
3. rt/$arch contains all the runtime architecture dependent code
4. rt/* contains the portable runtime code.

Obviously testing is needed. Specifically, while code is checked in for
the most popular architectures, it only has been tested on one computer!
Overall this is exciting and as been educational.

26 files changed, 213 insertions, 90 deletions

diff --git a/sys/rt/amd64/crt1.s b/sys/rt/amd64/crt1.s
deleted file mode 100644
index aaad0ba..0000000
--- a/sys/rt/amd64/crt1.s
+++ /dev/null
@@ -1,21 +0,0 @@
-global  _start
-
-; NOTE: assumes program loader has put argc, argv, envc, envp on stack
-section .text
-_start:
-    xor rbp,rbp    ; base pointer undefined: set to 0
-    mov r9,rdx     ; 6th arg: function to register with atexit()
-    pop rsi        ; 2nd arg: argc
-    mov rdx,rsp    ; 3rd arg: argv
-    and rsp,$-16   ; align stack pointer to 16 bytes
-    mov $_fini,r8  ; 5th arg: fini
-    mov $_init,rcx ; 4th arg: init
-    mov $main,rdi  ; 1st arg: main
-
-    call rt·boot   ; int boot(
-                   ;       int(*main)(int,char*[],char*[]),
-                   ;       int argc,
-                   ;       char *argv[],
-                   ;       init, fini, atexit);
-
-.loop: jmp .loop  ; should never reach...
diff --git a/sys/rt/amd64/crti.s b/sys/rt/amd64/crti.s
deleted file mode 100644
index eeba8b1..0000000
--- a/sys/rt/amd64/crti.s
+++ /dev/null
@@ -1,17 +0,0 @@
-global _init
-global _fini
-
-section .init
-_init:
-    ; call pushes an 8 byte return address on the stack
-    ; 64 bit ABI requires stack to be aligned to 16 bytes
-    ; we realign with another 8 byte address
-    push rbp
-    mov  rbp, rsp
-    ;; compiler inserts the rest here
-
-section .fini
-_fini:
-    push rbp ; see above
-    mov  rbp, rsp
-    ;; compiler inserts the rest here
diff --git a/sys/rt/amd64/crtn.s b/sys/rt/amd64/crtn.s
deleted file mode 100644
index 45d1a29..0000000
--- a/sys/rt/amd64/crtn.s
+++ /dev/null
@@ -1,9 +0,0 @@
-section .init
-    ;; compiler inserts here
-    pop rbp ;; undo our push
-    ret
-
-section .fini
-    ;; compiler inserts here
-    pop rbp ;; undo our push
-    ret
diff --git a/sys/rt/amd64/rt1.s b/sys/rt/amd64/rt1.s
new file mode 100644
index 0000000..afb5c32
--- /dev/null
+++ b/sys/rt/amd64/rt1.s
@@ -0,0 +1,9 @@
+.extern rt·thunk
+
+.text
+.global _start
+_start:
+	xor %rbp,%rbp
+	mov %rsp,%rdi
+	andq $-16,%rsp
+	call rt·thunk
diff --git a/sys/rt/amd64/rti.s b/sys/rt/amd64/rti.s
new file mode 100644
index 0000000..4788968
--- /dev/null
+++ b/sys/rt/amd64/rti.s
@@ -0,0 +1,9 @@
+.section .init
+.global _init
+_init:
+	push %rax
+
+.section .fini
+.global _fini
+_fini:
+	push %rax
diff --git a/sys/rt/amd64/rtn.s b/sys/rt/amd64/rtn.s
new file mode 100644
index 0000000..29198b7
--- /dev/null
+++ b/sys/rt/amd64/rtn.s
@@ -0,0 +1,7 @@
+.section .init
+	pop %rax
+	ret
+
+.section .fini
+	pop %rax
+	ret
diff --git a/sys/rt/arm/rt1.s b/sys/rt/arm/rt1.s
new file mode 100644
index 0000000..2adfee8
--- /dev/null
+++ b/sys/rt/arm/rt1.s
@@ -0,0 +1,11 @@
+.extern rt·thunk
+
+.text
+.global _start
+.type _start, %function
+_start:
+ 	mov fp, #0
+ 	mov lr, #0
+	mov a1, sp
+	and sp, sp, #-16
+ 	bl rt·thunk
diff --git a/sys/rt/arm/rti.s b/sys/rt/arm/rti.s
new file mode 100644
index 0000000..18dc1e4
--- /dev/null
+++ b/sys/rt/arm/rti.s
@@ -0,0 +1,13 @@
+.syntax unified
+
+.section .init
+.global _init
+.type _init,%function
+_init:
+	push {r0,lr}
+
+.section .fini
+.global _fini
+.type _fini,%function
+_fini:
+	push {r0,lr}
diff --git a/sys/rt/arm/rtn.s b/sys/rt/arm/rtn.s
new file mode 100644
index 0000000..dc020f9
--- /dev/null
+++ b/sys/rt/arm/rtn.s
@@ -0,0 +1,9 @@
+.syntax unified
+
+.section .init
+	pop {r0,lr}
+	bx lr
+
+.section .fini
+	pop {r0,lr}
+	bx lr
diff --git a/sys/rt/arm64/rt1.s b/sys/rt/arm64/rt1.s
new file mode 100644
index 0000000..d0f76da
--- /dev/null
+++ b/sys/rt/arm64/rt1.s
@@ -0,0 +1,10 @@
+ .extern rt·thunk
+
+.global _start
+.type _start, %function
+_start:
+	mov x29, #0
+	mov x30, #0
+	mov x0, sp
+	and sp, x0, #-16
+	b rt·thunk
diff --git a/sys/rt/arm64/rti.s b/sys/rt/arm64/rti.s
new file mode 100644
index 0000000..775df0a
--- /dev/null
+++ b/sys/rt/arm64/rti.s
@@ -0,0 +1,13 @@
+.section .init
+.global _init
+.type _init,%function
+_init:
+	stp x29,x30,[sp,-16]!
+	mov x29,sp
+
+.section .fini
+.global _fini
+.type _fini,%function
+_fini:
+	stp x29,x30,[sp,-16]!
+	mov x29,sp
diff --git a/sys/rt/arm64/rtn.s b/sys/rt/arm64/rtn.s
new file mode 100644
index 0000000..73cab69
--- /dev/null
+++ b/sys/rt/arm64/rtn.s
@@ -0,0 +1,7 @@
+.section .init
+	ldp x29,x30,[sp],#16
+	ret
+
+.section .fini
+	ldp x29,x30,[sp],#16
+	ret
diff --git a/sys/rt/boot.c b/sys/rt/boot.c
index 197e861..76f5599 100644
--- a/sys/rt/boot.c
+++ b/sys/rt/boot.c
@@ -1,34 +1,41 @@
+#include <u.h>
 #include <rt.h>
+#include <elf.h>
 
-int
-rt·boot(
-    int  (*main)(int,char **,char **), int argc, char **argv,
-    int  (*init)(int, char **, char **), void (*fini)(void),
-    void (*exit)(void)
-)
+/* tell linker to go find */
+int __libc_start_main(int (*main)(int,char **,char **), int argc, char **argv);
+
+#define NAUX 38
+static void
+rt·init(char **env, char *program)
 {
-    char **envp = argv+argc+1;
-    rt·environ = envp;
+	uintptr i, *auxv, aux[NAUX];
+	for(i=0; env[i]; i++)
+		;
+	rt·context.auxv = auxv = (void *)(env+i+1);
+	for(i=0; auxv[i]; i+=2)
+		if(auxv[i]<NAUX)
+			aux[auxv[i]] = auxv[i+1];
 
-    if(exit)
-        rt·context.exit = exit;
+	rt·context.sysinfo  = aux[AT_SYSINFO];
+	rt·context.pagesize = aux[AT_PAGESZ];
+}
 
-    if(fini)
-        rt·context.fini = fini;
+int
+rt·boot(int (*main)(), int argc, char **argv, void (*init)(), void (*fini)(), void (*exit)())
+{
+	char **env = argv+argc+1;
+	rt·init(env, argv[0]);
 
-    if(init){
-        rt·context.init = init;
-        init(argc, argv, envp);
-    }
+	/* ring libc, anyone home? */
+	__libc_start_main(main, argc, argv);
 
-    /* XXX:
-     * we could call __libc_start_main of musl here:
-     * this would give us the normal C runtime along with ours
-     * which would allow seamless linking to other libraries
-     * or we can implement a compatibility layer
-     */
-    rt·exit(main(argc, argv, envp));
+	/* no? ok we continue on if there is no libc linked */
+	init();
+	if(fini)
+		rt·context.fini = fini;
+	if(exit)
+		rt·context.exit = exit;
 
-    /* should never get here */
-    return 0;
+	rt·exit(main(argc, argv, env));
 }
diff --git a/sys/rt/context.c b/sys/rt/context.c
new file mode 100644
index 0000000..fb383e3
--- /dev/null
+++ b/sys/rt/context.c
@@ -0,0 +1,4 @@
+#include <u.h>
+#include <rt.h>
+
+struct rt·Context rt·context;
diff --git a/sys/rt/dummy.c b/sys/rt/dummy.c
new file mode 100644
index 0000000..f123d16
--- /dev/null
+++ b/sys/rt/dummy.c
@@ -0,0 +1,6 @@
+#include <u.h>
+#include <rt.h>
+
+/* provide a dummy implementation if libc is not linked */
+static int noop(int (*main)(), int argc, char **argv) { return 0; }
+weakalias(noop, __libc_start_main);
diff --git a/sys/rt/env.c b/sys/rt/env.c
index ddd559d..be710ae 100644
--- a/sys/rt/env.c
+++ b/sys/rt/env.c
@@ -1,3 +1,4 @@
+#include <u.h>
 #include <rt.h>
 
-char *rt·environ = 0;
+char **rt·environ = 0;
diff --git a/sys/rt/exit.c b/sys/rt/exit.c
index 55a684a..e6027b7 100644
--- a/sys/rt/exit.c
+++ b/sys/rt/exit.c
@@ -1,15 +1,27 @@
+#include <u.h>
 #include <rt.h>
-#include <sys.h>
+#include <syscall.h>
 
-void
-rt·exit(int code)
+/* XXX:
+ * if we are here, we are in charge, call exit syscalls
+ * think of better way to encapsulate these syscalls?
+ */
+static noreturn void
+rt·shutdown(int code)
 {
     if(rt·context.fini)
         rt·context.fini();
     if(rt·context.exit)
         rt·context.exit();
 
-    /* call exit syscalls here */
+	_syscall1(·ExitGroup, code);
+    for(;;)
+		_syscall1(·Exit, code);
+}
+weakalias(rt·shutdown, exit);
 
-    for(;;);
+noreturn void
+rt·exit(int code)
+{
+	rt·shutdown(code);
 }
diff --git a/sys/rt/i386/rt1.s b/sys/rt/i386/rt1.s
new file mode 100644
index 0000000..aef4f05
--- /dev/null
+++ b/sys/rt/i386/rt1.s
@@ -0,0 +1,14 @@
+.extern rt·thunk
+
+.text
+.global _start
+_start:
+	xor %ebp,%ebp
+	mov %esp,%eax
+	and $-16,%esp
+	push %eax
+	push %eax
+	push %eax
+	push %eax
+	call rt·thunk
+");
diff --git a/sys/rt/i386/rti.s b/sys/rt/i386/rti.s
new file mode 100644
index 0000000..d2682a2
--- /dev/null
+++ b/sys/rt/i386/rti.s
@@ -0,0 +1,9 @@
+.section .init
+.global _init
+_init:
+	sub $12,%esp
+
+.section .fini
+.global _fini
+_fini:
+	sub $12,%esp
diff --git a/sys/rt/i386/rtn.s b/sys/rt/i386/rtn.s
new file mode 100644
index 0000000..f3b61e0
--- /dev/null
+++ b/sys/rt/i386/rtn.s
@@ -0,0 +1,7 @@
+.section .init
+	add $12,%esp
+	ret
+
+.section .fini
+	add $12,%esp
+	ret
diff --git a/sys/rt/include/rt.h b/sys/rt/include/rt.h
deleted file mode 100644
index dd1598b..0000000
--- a/sys/rt/include/rt.h
+++ /dev/null
@@ -1,12 +0,0 @@
-#pragma once
-
-extern struct rt·context
-{
-    void (*init)(void);
-    void (*fini)(void);
-    void (*exit)(void);
-} rt·context;
-
-extern char **rt·environ;
-
-void rt·exit(int code);
diff --git a/sys/rt/riscv64/rt1.s b/sys/rt/riscv64/rt1.s
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/sys/rt/riscv64/rt1.s
diff --git a/sys/rt/riscv64/rti.s b/sys/rt/riscv64/rti.s
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/sys/rt/riscv64/rti.s
diff --git a/sys/rt/riscv64/rtn.s b/sys/rt/riscv64/rtn.s
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/sys/rt/riscv64/rtn.s
diff --git a/sys/rt/stack.c b/sys/rt/stack.c
new file mode 100644
index 0000000..5ab7a8a
--- /dev/null
+++ b/sys/rt/stack.c
@@ -0,0 +1,17 @@
+#include <u.h>
+#include <arch/atomic.h>
+
+/* unprefixed names determined by gcc */
+uintptr __stack_chk_guard;
+
+void
+rt·guardstack(void)
+{
+	__stack_chk_guard = (uintptr)&__stack_chk_guard*1103515245;
+}
+
+void
+__stack_chk_fail(void)
+{
+	atomic·crash();
+}
diff --git a/sys/rt/thunk.c b/sys/rt/thunk.c
new file mode 100644
index 0000000..3e9fa35
--- /dev/null
+++ b/sys/rt/thunk.c
@@ -0,0 +1,17 @@
+#include <u.h>
+#include <rt.h>
+
+int  main();
+void _init() __attribute__((weak));
+void _fini() __attribute__((weak));
+
+void rt·boot(int (*)(), int , char **, void (*)(), void (*)(), void (*)());
+
+void
+rt·thunk(intptr *stack)
+{
+	int argc	= stack[0];
+	char **argv = (void *)(stack+1);
+
+	rt·boot(main, argc, argv, _init, _fini, 0);
+}