Get GDB working

You can now say `gdb hello.com.dbg` and it'll work perfectly.

libc/runtime/clone-linux.S

@@ -30,18 +30,18 @@
 //	@param	8(rsp)	x6	is arg
 //	@return	tid of child on success, or -errno on error
 sys_clone_linux:
+	beg
+	pro
 #ifdef __x86_64__
-	push	%rbp
-	mov	%rsp,%rbp
-	push	%rbx
+	cpush	%rbx
 	mov	%rcx,%r10
 	mov	16(%rbp),%rbx
 	mov	$56,%eax		// __NR_clone
 	syscall
 	test	%rax,%rax
 	jz	2f
-0:	pop	%rbx
-	pop	%rbp
+0:	cpop	%rbx
+	epi
 	ret
 2:	xor	%ebp,%ebp		// child thread
 	mov	%rbx,%rdi		// arg
@@ -50,15 +50,13 @@ sys_clone_linux:
 	mov	$60,%eax		// __NR_exit(exitcode)
 	syscall
 #elif defined(__aarch64__)
-	stp	x29,x30,[sp,#-16]!
-	mov	x29,sp
 	mov	x8,x3			// swap x3 and x4
 	mov	x3,x4			// swap x3 and x4
 	mov	x4,x8			// swap x3 and x4
 	mov	x8,#220			// __NR_clone
 	svc	#0
 	cbz	x0,2f
-	ldp	x29,x30,[sp],#16
+	epi
 	ret
 2:	mov	x29,#0			// wipe backtrace
 	mov	x28,x3			// set cosmo tls
@@ -69,4 +67,5 @@ sys_clone_linux:
 #else
 #error "unsupported architecture"
 #endif
+	end
 	.endfn	sys_clone_linux,globl,hidden

libc/runtime/cosmo.S

@@ -32,8 +32,8 @@
 //	@param	rdx is environ
 //	@param	rcx is auxv
 //	@noreturn
-cosmo:	push	%rbp
-	mov	%rsp,%rbp
+cosmo:	beg
+	pro
 	mov	%edi,%r12d
 	mov	%rsi,%r13
 	mov	%rdx,%r14
@@ -104,7 +104,10 @@ cosmo:	push	%rbp
 	je	2f
 	push	%rax
 	push	%rax
-	call	.Largs
+	mov	%r12d,%edi
+	mov	%r13,%rsi
+	mov	%r14,%rdx
+	mov	%r15,%rcx
 	call	*(%rax)
 	pop	%rax
 	pop	%rax
@@ -112,17 +115,15 @@ cosmo:	push	%rbp
 	jmp	1b
 
 //	call main()
-2:	call	.Largs
+2:	mov	%r12d,%edi
+	mov	%r13,%rsi
+	mov	%r14,%rdx
+	mov	%r15,%rcx
 	.weak	main
 	call	main
 	xchg	%eax,%edi
 	call	exit
-
-.Largs:	mov	%r12d,%edi
-	mov	%r13,%rsi
-	mov	%r14,%rdx
-	mov	%r15,%rcx
-	ret
+	end
 	.endfn	cosmo,weak
 
 //	Enables Thread Local Storage.

libc/runtime/ftrace-hook.S

@@ -28,8 +28,12 @@ ftrace_hook:
 
 	cmpl	$0,__ftrace(%rip)
 	jle	1f
+	.cfi_startproc
 	push	%rbp
+	.cfi_def_cfa_offset 16
+	.cfi_offset %rbp, -16
 	mov	%rsp,%rbp
+	.cfi_def_cfa_register %rbp
 	and	$-16,%rsp
 	sub	$128,%rsp
 	movdqu	%xmm0,-0x80(%rbp)
@@ -41,13 +45,21 @@ ftrace_hook:
 	movdqu	%xmm6,-0x20(%rbp)
 	movdqu	%xmm7,-0x10(%rbp)
 	push	%rax
+	.cfi_offset %rax, -24
 	push	%rcx
+	.cfi_offset %rcx, -32
 	push	%rdx
+	.cfi_offset %rdx, -40
 	push	%rdi
+	.cfi_offset %rdi, -48
 	push	%rsi
+	.cfi_offset %rsi, -56
 	push	%r8
+	.cfi_offset %r8, -64
 	push	%r9
+	.cfi_offset %r9, -72
 	push	%r10
+	.cfi_offset %r10, -80
 	call	ftracer
 	movdqu	-0x80(%rbp),%xmm0
 	movdqu	-0x70(%rbp),%xmm1
@@ -66,12 +78,20 @@ ftrace_hook:
 	pop	%rcx
 	pop	%rax
 	leave
+	.cfi_restore %rbp
+	.cfi_def_cfa %rsp, 8
 1:	ret
+	.cfi_endproc
 
 #elif defined(__aarch64__)
 
 	stp	x29,x30,[sp,-384]!
+	.cfi_startproc
+	.cfi_def_cfa_offset 384
+	.cfi_offset 29, -384	// x29 (fp) is saved at [sp - 384]
+	.cfi_offset 30, -376	// x30 (lr) is saved at [sp - 376]
 	mov	x29,sp
+	.cfi_def_cfa_register 29
 	stp	x0,x1,[sp,16]
 
 	adrp	x0,__ftrace
@@ -80,18 +100,45 @@ ftrace_hook:
 	ble	1f
 
 	stp	x2,x3,[sp,32]
+	.cfi_offset 2, -352
+	.cfi_offset 3, -344
 	stp	x4,x5,[sp,48]
+	.cfi_offset 4, -336
+	.cfi_offset 5, -328
 	stp	x6,x7,[sp,64]
+	.cfi_offset 6, -320
+	.cfi_offset 7, -312
 	stp	x8,x9,[sp,80]
+	.cfi_offset 8, -304
+	.cfi_offset 9, -296
 	stp	x10,x11,[sp,96]
+	.cfi_offset 10, -288
+	.cfi_offset 11, -280
 	stp	x12,x13,[sp,112]
+	.cfi_offset 12, -272
+	.cfi_offset 13, -264
 	stp	x14,x15,[sp,128]
+	.cfi_offset 14, -256
+	.cfi_offset 15, -248
 	stp	x16,x19,[sp,160]
+	.cfi_offset 16, -224
+	.cfi_offset 19, -216
 	stp	x20,x21,[sp,176]
+	.cfi_offset 20, -208
+	.cfi_offset 21, -200
 	stp	x22,x23,[sp,192]
+	.cfi_offset 22, -192
+	.cfi_offset 23, -184
 	stp	x24,x25,[sp,208]
+	.cfi_offset 24, -176
+	.cfi_offset 25, -168
 	stp	x26,x27,[sp,224]
+	.cfi_offset 26, -160
+	.cfi_offset 27, -152
 	stp	x17,x28,[sp,240]
+	.cfi_offset 17, -144
+	.cfi_offset 28, -136
+	// No CFI directives needed for FP registers
 	stp	q0,q1,[sp,256]
 	stp	q2,q3,[sp,288]
 	stp	q4,q5,[sp,320]
@@ -119,7 +166,12 @@ ftrace_hook:
 
 1:	ldp	x0,x1,[sp,16]
 	ldp	x29,x30,[sp],384
+	.cfi_restore 29
+	.cfi_restore 30
+	.cfi_def_cfa 7, 0	// On some ARM systems the stack pointer is represented by register 7
+	.cfi_def_cfa_offset 0
 	ret
+	.cfi_endproc
 
 #endif /* __x86_64__ */
 	.endfn	ftrace_hook,globl,hidden