diff --git a/ape/aarch64.lds b/ape/aarch64.lds
index 0a232a2da..48562d2a2 100644
--- a/ape/aarch64.lds
+++ b/ape/aarch64.lds
@@ -259,6 +259,9 @@ SECTIONS {
   .debug_ranges       0 : { *(.debug_ranges) }
   .debug_macro        0 : { *(.debug_macro) }
   .debug_addr         0 : { *(.debug_addr) }
+  .debug_names        0 : { *(.debug_names) }
+  .debug_loclists     0 : { *(.debug_loclists) }
+  .debug_str_offsets  0 : { *(.debug_str_offsets) }
   .ARM.attributes     0 : { KEEP(*(.ARM.attributes)) KEEP(*(.gnu.attributes)) }
   .note.gnu.arm.ident 0 : { KEEP(*(.note.gnu.arm.ident)) }
 
diff --git a/ape/ape.lds b/ape/ape.lds
index 155b0aad9..ac82bde00 100644
--- a/ape/ape.lds
+++ b/ape/ape.lds
@@ -386,6 +386,13 @@ SECTIONS {
     _tbss_end = .;
   } :Tls
 
+  .eh_frame : {
+    __eh_frame_start = .;
+    KEEP(*(.eh_frame))
+    *(.eh_frame.*)
+    __eh_frame_end = .;
+  } :Ram
+
   .data . : {
 /*BEGIN: Read/Write Data */
 #if SupportsWindows()
@@ -426,11 +433,6 @@ SECTIONS {
     KEEP(*(.dtors))
     __fini_array_end = .;
 
-    __eh_frame_start = .;
-    KEEP(*(.eh_frame))
-    *(.eh_frame.*)
-    __eh_frame_end = .;
-
 /*BEGIN: Post-Initialization Read-Only */
     . = ALIGN(. != 0 ? __SIZEOF_POINTER__ : 0);
     KEEP(*(SORT_BY_NAME(.piro.relo.sort.*)))
@@ -439,7 +441,6 @@ SECTIONS {
     KEEP(*(.piro.pad.data))
     *(.igot.plt)
     KEEP(*(.dataepilogue))
-
     . = ALIGN(. != 0 ? CONSTANT(COMMONPAGESIZE) : 0);
 /*END: NT FORK COPYING */
     _edata = .;
@@ -519,6 +520,9 @@ SECTIONS {
   .debug_rnglists    0 : { *(.debug_rnglists) }
   .debug_macro       0 : { *(.debug_macro) }
   .debug_addr        0 : { *(.debug_addr) }
+  .debug_names       0 : { *(.debug_names) }
+  .debug_loclists    0 : { *(.debug_loclists) }
+  .debug_str_offsets 0 : { *(.debug_str_offsets) }
   .gnu.attributes    0 : { KEEP(*(.gnu.attributes)) }
   .GCC.command.line  0 : { *(.GCC.command.line) }
 
@@ -582,11 +586,11 @@ ape_rom_memsz = ape_rom_filesz;
 ape_rom_align = CONSTANT(COMMONPAGESIZE);
 ape_rom_rva = RVA(ape_rom_vaddr);
 
-ape_ram_vaddr = ADDR(.data);
+ape_ram_vaddr = ADDR(.eh_frame);
 ape_ram_offset = ape_ram_vaddr - __executable_start;
-ape_ram_paddr = LOADADDR(.data);
-ape_ram_filesz = ADDR(.bss) - ADDR(.data);
-ape_ram_memsz = _end - ADDR(.data);
+ape_ram_paddr = LOADADDR(.eh_frame);
+ape_ram_filesz = ADDR(.bss) - ADDR(.eh_frame);
+ape_ram_memsz = _end - ADDR(.eh_frame);
 ape_ram_align = CONSTANT(COMMONPAGESIZE);
 ape_ram_rva = RVA(ape_ram_vaddr);
 
diff --git a/libc/calls/getntsyspath.S b/libc/calls/getntsyspath.S
index b8f2b65cd..bd8178bd3 100644
--- a/libc/calls/getntsyspath.S
+++ b/libc/calls/getntsyspath.S
@@ -28,8 +28,8 @@
 //	@return	rdi is rdi+edx
 	.text.startup
 __getntsyspath:
-	push	%rbp
-	mov	%rsp,%rbp
+	beg
+	pro
 	push	%rdx
 	movpp	%rdi,%rcx		# call f=%rax(p1=%rcx,p2=%rdx)
 	sub	$40,%rsp
@@ -55,6 +55,7 @@ __getntsyspath:
 	jne	2f
 	movb	$'/',-1(%rdi)
 2:	.loop	1b
-	leave
+	epi
 	ret
+	end
 	.endfn	__getntsyspath,globl,hidden
diff --git a/libc/crt/crt.S b/libc/crt/crt.S
index d34bc83b8..74226c641 100644
--- a/libc/crt/crt.S
+++ b/libc/crt/crt.S
@@ -47,7 +47,14 @@ __oops_win32:
 //	@note	ape.S and ape-loader both set RCX to XNU on Darwin
 //	@noreturn
 _start:
-#ifdef __x86_64__
+	.cfi_startproc
+#if defined(__x86_64__)
+	.cfi_undefined rip
+#elif defined(__aarch64__)
+	.cfi_undefined x30
+#endif /* __x86_64__ */
+
+#if defined(__x86_64__)
 
 #if SupportsFreebsd()
 //	detect free besiyata dishmaya
@@ -159,4 +166,5 @@ _start:
 #else
 #error "architecture unsupported"
 #endif /* __x86_64__ */
+	.cfi_endproc
 	.endfn	_start,weak,hidden
diff --git a/libc/intrin/cosmo_futex_thunk.S b/libc/intrin/cosmo_futex_thunk.S
index 1ce0d5917..ad65cc106 100644
--- a/libc/intrin/cosmo_futex_thunk.S
+++ b/libc/intrin/cosmo_futex_thunk.S
@@ -21,16 +21,15 @@
 .privileged
 
 cosmo_futex_thunk:
+	beg
+	pro
 #ifdef __x86_64__
-	push	%rbp
-	mov	%rsp,%rbp
 	mov	%rcx,%r10
 	mov	__NR_futex,%eax
 	clc
 	syscall
 	jnc	1f
 	neg	%eax
-1:	pop	%rbp
 #elif defined(__aarch64__)
 	ldr	x7,=__hostos
 	ldr	w7,[x7]
@@ -46,5 +45,7 @@ cosmo_futex_thunk:
 #else
 #error "unsupported architecture"
 #endif /* __x86_64__ */
-1:	ret
+1:	epi
+	ret
+	end
 	.endfn	cosmo_futex_thunk,globl,hidden
diff --git a/libc/intrin/getcontext.S b/libc/intrin/getcontext.S
index 8be4f58eb..bdfaded97 100644
--- a/libc/intrin/getcontext.S
+++ b/libc/intrin/getcontext.S
@@ -26,7 +26,9 @@
 //	@see	setcontext()
 	.ftrace1
 getcontext:
+	beg
 	.ftrace2
 #include "libc/intrin/getcontext.inc"
 	jmp	__getcontextsig
+	end
 	.endfn	getcontext,globl
diff --git a/libc/intrin/swapcontext.S b/libc/intrin/swapcontext.S
index 6d2e517e6..b40b86777 100644
--- a/libc/intrin/swapcontext.S
+++ b/libc/intrin/swapcontext.S
@@ -31,17 +31,17 @@
 //	@returnstwice
 	.ftrace1
 swapcontext:
+	beg
 	.ftrace2
 #include "libc/intrin/getcontext.inc"
 #ifdef __x86_64__
-	push	%rbp
-	mov	%rsp,%rbp
-	push	%rsi
-	push	%rsi
+	pro
+	cpush	%rsi
+	cpush	%rsi
 	call	__swapcontextsig
-	pop	%rdi
-	pop	%rdi
-	pop	%rbp
+	cpop	%rdi
+	cpop	%rdi
+	epi
 	test	%eax,%eax
 	jnz	1f
 #elif defined(__aarch64__)
@@ -56,4 +56,5 @@ swapcontext:
 #endif
 	jmp	__tailcontext
 1:	ret
+	end
 	.endfn	swapcontext,globl
diff --git a/libc/intrin/sys_sched_yield.S b/libc/intrin/sys_sched_yield.S
index 2bfaa7ccb..f78f48712 100644
--- a/libc/intrin/sys_sched_yield.S
+++ b/libc/intrin/sys_sched_yield.S
@@ -24,9 +24,9 @@
 //
 //	@return	0 on success, or -1 w/ errno
 sys_sched_yield:
+	beg
 #ifdef __x86_64__
-	push	%rbp
-	mov	%rsp,%rbp
+	pro
 	xor	%eax,%eax
 	mov	__hostos(%rip),%dl
 
@@ -84,13 +84,16 @@ sys_sched_yield:
 //	fails a positive or negative errno might get returned.
 #endif
 
-9:	leave
+9:	epi
 	ret
 
 #elif defined(__aarch64__)
 
 	stp	x29,x30,[sp,-32]!
 	mov	x29,sp
+	.cfi_adjust_cfa_offset 32
+	.cfi_rel_offset x29,16
+	.cfi_rel_offset x30,24
 	mov	x3,0
 	mov	x2,0
 	add	x4,sp,16
@@ -101,10 +104,14 @@ sys_sched_yield:
 	mov	x16,#0x5d			// select(0,0,0,0,&blah) for xnu
 	svc	0
 	ldp	x29,x30,[sp],32
+	.cfi_adjust_cfa_offset -32
+	.cfi_restore x30
+	.cfi_restore x29
 	ret
 
 #else
 #error "arch unsupported"
 #endif
+	end
 	.endfn	sys_sched_yield,globl
 	.previous
diff --git a/libc/macros.h b/libc/macros.h
index 9a29e396a..257007a84 100644
--- a/libc/macros.h
+++ b/libc/macros.h
@@ -158,6 +158,60 @@
 	.weak	\canonical
 .endm
 
+.macro	beg
+	.cfi_startproc
+.endm
+
+.macro	pro
+#if defined(__x86_64__)
+	push	%rbp
+	.cfi_adjust_cfa_offset 8
+	.cfi_rel_offset %rbp,0
+	mov	%rsp,%rbp
+	.cfi_def_cfa_register %rbp
+#elif defined(__aarch64__)
+	stp	x29,x30,[sp,-16]!
+	mov	x29,sp
+	.cfi_adjust_cfa_offset 16
+	.cfi_rel_offset x29,0
+	.cfi_rel_offset x30,8
+#else
+#error "unsupported architecture"
+#endif
+.endm
+
+.macro	epi
+#if defined(__x86_64__)
+	.cfi_def_cfa_register %rsp
+	leave
+	.cfi_adjust_cfa_offset -8
+	.cfi_restore %rbp
+#elif defined(__aarch64__)
+	ldp	x29,x30,[sp],#16
+	.cfi_adjust_cfa_offset -16
+	.cfi_restore x30
+	.cfi_restore x29
+#else
+#error "unsupported architecture"
+#endif
+.endm
+
+.macro	end
+	.cfi_endproc
+.endm
+
+.macro	cpush	reg:req
+	push	\reg
+	.cfi_adjust_cfa_offset 8
+	.cfi_rel_offset \reg,0
+.endm
+
+.macro	cpop	reg:req
+	pop	\reg
+	.cfi_adjust_cfa_offset -8
+	.cfi_restore \reg
+.endm
+
 #ifdef __aarch64__
 .macro	jmp	dest:req
 	b	\dest
diff --git a/libc/nexgen32e/djbsort-avx2.S b/libc/nexgen32e/djbsort-avx2.S
index 8f51d678e..70868472d 100644
--- a/libc/nexgen32e/djbsort-avx2.S
+++ b/libc/nexgen32e/djbsort-avx2.S
@@ -7,8 +7,8 @@
 //	@note	public domain
 //	@see	en.wikipedia.org/wiki/Sorting_network
 djbsort_avx2:
-	push	%rbp
-	mov	%rsp,%rbp
+	beg
+	pro
 	push	%r15
 	push	%r14
 	push	%r13
@@ -795,11 +795,13 @@ djbsort_avx2:
 	pop	%r13
 	pop	%r14
 	pop	%r15
-	pop	%rbp
+	epi
 	ret
+	end
 	.endfn	djbsort_avx2,globl,hidden
 
 minmax_vector:
+	beg
 	cmp	$7,%rdx
 	jg	.L13
 .L2:	test	%rdx,%rdx
@@ -838,9 +840,11 @@ minmax_vector:
 	sub	$8,%rdx
 	jne	.L7
 	ret
+	end
 	.endfn	minmax_vector
 
 int32_twostages_32:
+	beg
 	sub	$-128,%rdi
 .L17:	lea	-128(%rdi),%rax
 	test	%rsi,%rsi
@@ -866,13 +870,14 @@ int32_twostages_32:
 	add	$512,%rdi
 	jmp	.L17
 .L21:	ret
+	end
 	.endfn	int32_twostages_32
 
 int32_threestages:
-	push	%rbp
+	beg
+	pro
 	imul	$-24,%rdx,%r8
 	lea	0(,%rdx,8),%rax
-	mov	%rsp,%rbp
 	push	%r15
 	push	%r14
 	push	%r13
@@ -961,11 +966,13 @@ int32_threestages:
 	pop	%r13
 	pop	%r14
 	pop	%r15
-	pop	%rbp
+	epi
 	ret
+	end
 	.endfn	int32_threestages
 
 merge16_finish:
+	beg
 	vpminsd	%ymm1,%ymm0,%ymm3
 	vpmaxsd	%ymm1,%ymm0,%ymm0
 	vperm2i128 $32,%ymm0,%ymm3,%ymm2
@@ -994,9 +1001,11 @@ merge16_finish:
 .L31:	vmovdqu	%ymm2,(%rdi)
 	vmovdqu	%ymm0,32(%rdi)
 	ret
+	end
 	.endfn	merge16_finish
 
 int32_sort_2power:
+	beg
 	push	%r13
 	lea	16(%rsp),%r13
 	andq	$-32,%rsp
@@ -2075,6 +2084,7 @@ int32_sort_2power:
 	lea	-16(%r13),%rsp
 	pop	%r13
 	ret
+	end
 	.endfn	int32_sort_2power
 
 	.rodata.cst32
diff --git a/libc/nexgen32e/gc.S b/libc/nexgen32e/gc.S
index 8dd47a41d..1e6f30266 100644
--- a/libc/nexgen32e/gc.S
+++ b/libc/nexgen32e/gc.S
@@ -32,7 +32,8 @@
 //	@param	rax,rdx,xmm0,xmm1,st0,st1 is return value
 //	@see	test/libc/runtime/gc_test.c
 	.ftrace1
-__gc:	.ftrace2
+__gc:	beg
+	.ftrace2
 
 #ifdef __x86_64__
 
@@ -47,8 +48,7 @@ __gc:	.ftrace2
 	mov	8(%r8),%r9
 	mov	16(%r8),%rdi
 	push	24(%r8)
-	push	%rbp
-	mov	%rsp,%rbp
+	pro
 	sub	$32,%rsp
 	mov	%rax,-8(%rbp)
 	mov	%rdx,-16(%rbp)
@@ -57,7 +57,7 @@ __gc:	.ftrace2
 	movdqa	-32(%rbp),%xmm0
 	mov	-16(%rbp),%rdx
 	mov	-8(%rbp),%rax
-	leave
+	epi
 	ret
 9:	ud2
 	nop
@@ -102,4 +102,5 @@ __gc:	.ftrace2
 
 #endif /* __x86_64__ */
 
+	end
 	.endfn	__gc,globl,hidden
diff --git a/libc/nexgen32e/gclongjmp.S b/libc/nexgen32e/gclongjmp.S
index 88f534e10..51f93cb15 100644
--- a/libc/nexgen32e/gclongjmp.S
+++ b/libc/nexgen32e/gclongjmp.S
@@ -31,7 +31,9 @@
 //	@noreturn
 	.ftrace1
 gclongjmp:
+	beg
 	.ftrace2
+	pro
 #ifdef __x86_64__
 	push	%rbp
 	mov	%rsp,%rbp
@@ -65,4 +67,5 @@ gclongjmp:
 #else
 #error "unsupported architecture"
 #endif /* __x86_64__ */
+	end
 	.endfn	gclongjmp,globl
diff --git a/libc/nexgen32e/longjmp.S b/libc/nexgen32e/longjmp.S
index aa4e0cfc7..5aefd029f 100644
--- a/libc/nexgen32e/longjmp.S
+++ b/libc/nexgen32e/longjmp.S
@@ -26,7 +26,7 @@
 //	@see	gclongjmp()
 //	@see	siglongjmp()
 	.ftrace1
-longjmp:
+longjmp:beg
 	.ftrace2
 _longjmp:
 #ifdef __x86_64__
@@ -61,6 +61,7 @@ _longjmp:
 #else
 #error "unsupported architecture"
 #endif
+	end
 	.endfn	longjmp,globl
 	.endfn	_longjmp,globl
 	.alias	longjmp,siglongjmp
diff --git a/libc/nexgen32e/nt2sysv.S b/libc/nexgen32e/nt2sysv.S
index e4461d1bb..185687de6 100644
--- a/libc/nexgen32e/nt2sysv.S
+++ b/libc/nexgen32e/nt2sysv.S
@@ -30,8 +30,8 @@
 //	@note	slower than __sysv2nt
 //	@see	NT2SYSV() macro
 __nt2sysv:
-	push	%rbp
-	mov	%rsp,%rbp
+	beg
+	pro
 	sub	$256,%rsp
 	push	%rbx
 	push	%rdi
@@ -48,6 +48,7 @@ __nt2sysv:
 	pop	%rsi
 	pop	%rdi
 	pop	%rbx
-	leave
+	epi
 	ret
+	end
 	.endfn	__nt2sysv,globl,hidden
diff --git a/libc/runtime/clone-linux.S b/libc/runtime/clone-linux.S
index 2c3a0caed..909d525fe 100644
--- a/libc/runtime/clone-linux.S
+++ b/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
diff --git a/libc/runtime/cosmo.S b/libc/runtime/cosmo.S
index 0b52e57d6..59fe944c1 100644
--- a/libc/runtime/cosmo.S
+++ b/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.
diff --git a/libc/runtime/ftrace-hook.S b/libc/runtime/ftrace-hook.S
index 56b66704c..cd25a18c4 100644
--- a/libc/runtime/ftrace-hook.S
+++ b/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
diff --git a/libc/sysv/systemfive.S b/libc/sysv/systemfive.S
index 76075a927..178892482 100644
--- a/libc/sysv/systemfive.S
+++ b/libc/sysv/systemfive.S
@@ -102,8 +102,8 @@ __pid:	.quad	0
 	.previous
 
 systemfive_cp:
-	push	%rbp
-	mov	%rsp,%rbp		// so backtraces work
+	beg
+	pro
 systemfive_cancellable:			// our pthread_cancel() miracle code
 	cmpb	$0,__tls_enabled(%rip)	// inspired by the musl libc design!
 	je	1f			// we handle linux and bsd together!
@@ -123,7 +123,7 @@ systemfive_cancellable:			// our pthread_cancel() miracle code
 	clc				// no cancellable system calls exist
 	syscall				// that have 7+ args on the bsd OSes
 systemfive_cancellable_end:		// i/o calls park here for long time
-	pop	%rbp
+	epi
 	jnc	2f
 	neg	%rax			// turns bsd errno to system v errno
 2:	cmp	$-4095,%rax		// but we still check again on eintr
@@ -144,11 +144,13 @@ systemfive_cancellable_end:		// i/o calls park here for long time
 	je	systemfive_errno	// we aren't actually cancelled
 	jmp	4f			// now we are in fact cancelled
 systemfive_cancel:			// SIGTHR will jump here too
-	pop	%rbp
+	epi
 4:	jmp	_pthread_cancel_ack	// tail call
 	.weak	_pthread_cancel_ack	// must be linked if we're cancelled
+	end
 #if IsModeDbg()
 not_a_cancellation_point:		// need BEGIN/END_CANCELLATION_POINT
+	beg
 	nop
 	.weak	report_cancellation_point
 5:	ezlea	report_cancellation_point,cx
@@ -157,6 +159,7 @@ not_a_cancellation_point:		// need BEGIN/END_CANCELLATION_POINT
 	call	*%rcx
 6:	ud2
 	nop
+	end
 #endif
 	.globl	systemfive_cancellable_end
 	.globl	systemfive_cancellable
@@ -166,19 +169,20 @@ not_a_cancellation_point:		// need BEGIN/END_CANCELLATION_POINT
 .Lanchorpoint:
 #if SupportsLinux() || SupportsMetal()
 systemfive_linux:
+	beg
 	and	$0xfff,%eax		// remove nonlinux bits from ordinal
 	cmp	$0xfff,%eax		// checks if unsupported by platform
 	je	systemfive_enosys	// never taken branches cost nothing
 	btr	$11,%eax		// 0x800 means a call is cancellable
 	jc	systemfive_cp		// it is handled by the holiest code
 	mov	%rcx,%r10		// syscall instruction clobbers %rcx
-	push	%rbp			// linux never reads args from stack
-	mov	%rsp,%rbp		// having frame will help backtraces
+	pro				// linux never reads args from stack
 	syscall				// this is known as a context switch
-	pop	%rbp			// next we check to see if it failed
+	epi				// next we check to see if it failed
 	cmp	$-4095,%rax		// system five nexgen32e abi § a.2.1
 	jae	systemfive_error	// encodes errno as neg return value
 	ret
+	end
 	.endfn	systemfive_linux,globl,hidden
 systemfive_error:
 	neg	%eax
@@ -186,27 +190,35 @@ systemfive_error:
 	.endfn	systemfive_error,globl,hidden
 #endif
 systemfive_errno:
+	beg
 	xchg	%eax,%ecx
 	call	__errno_location
 	mov	%ecx,(%rax)		// normalize to c library convention
 	push	$-1			// negative one is only error result
 	pop	%rax			// the push pop is to save code size
 	ret
+	end
 	.endfn	systemfive_errno,globl,hidden
 systemfive_enosys:
+	beg
 	mov	ENOSYS(%rip),%eax
 	jmp	systemfive_errno
+	end
 	.endfn	systemfive_enosys,globl,hidden
 #if SupportsNetbsd()
 systemfive_netbsd:
+	beg
 	shr	$4*13,%rax
 	jmp	systemfive_bsdscrub
+	end
 	.endfn	systemfive_netbsd,globl,hidden
 #endif
 #if SupportsOpenbsd()
 systemfive_openbsd:
+	beg
 	shr	$4*10,%rax
 	jmp	systemfive_bsdscrub
+	end
 	.endfn	systemfive_openbsd,globl,hidden
 #endif
 #if SupportsFreebsd()
@@ -222,6 +234,7 @@ systemfive_bsdscrub:
 //	𝑠𝑙𝑖𝑑𝑒
 	.endfn	systemfive_bsdscrub,globl,hidden
 systemfive_bsd:
+	beg
 	cmp	$0xfff,%ax
 	je	systemfive_enosys
 	btr	$11,%eax		// checks/reset the 800 cancellable bit
@@ -230,6 +243,7 @@ systemfive_bsd:
 	syscall				// bsd will need arg on stack sometimes
 	jc	systemfive_errno	// bsd sets carry flag if %rax is errno
 	ret
+	end
 	.endfn	systemfive_bsd
 #endif
 #if SupportsXnu()