From: David Benjamin <davidben@google.com>
Date: Tue, 29 Jan 2019 05:12:15 +0000 (+0000)
Subject: Fix some CFI issues in x86_64 assembly
X-Git-Tag: OpenSSL_1_1_1b~28
X-Git-Url: https://git.librecmc.org/?a=commitdiff_plain;h=2086edb799acf6ad5ef0bb53aa3b17abf4f7f992;p=oweals%2Fopenssl.git

Fix some CFI issues in x86_64 assembly

The add/double shortcut in ecp_nistz256-x86_64.pl left one instruction
point that did not unwind, and the "slow" path in AES_cbc_encrypt was
not annotated correctly. For the latter, add
.cfi_{remember,restore}_state support to perlasm.

Next, fill in a bunch of functions that are missing no-op .cfi_startproc
and .cfi_endproc blocks. libunwind cannot unwind those stack frames
otherwise.

Finally, work around a bug in libunwind by not encoding rflags. (rflags
isn't a callee-saved register, so there's not much need to annotate it
anyway.)

These were found as part of ABI testing work in BoringSSL.

Reviewed-by: Richard Levitte <levitte@openssl.org>
GH: #8109
(cherry picked from commit c0e8e5007ba5234d4d448e82a1567e0c4467e629)
---

diff --git a/crypto/aes/asm/aes-x86_64.pl b/crypto/aes/asm/aes-x86_64.pl
index 4d1dc9c701..55c64147cb 100755
--- a/crypto/aes/asm/aes-x86_64.pl
+++ b/crypto/aes/asm/aes-x86_64.pl
@@ -554,6 +554,7 @@ $code.=<<___;
 .type	_x86_64_AES_encrypt_compact,\@abi-omnipotent
 .align	16
 _x86_64_AES_encrypt_compact:
+.cfi_startproc
 	lea	128($sbox),$inp			# size optimization
 	mov	0-128($inp),$acc1		# prefetch Te4
 	mov	32-128($inp),$acc2
@@ -587,6 +588,7 @@ $code.=<<___;
 	xor	8($key),$s2
 	xor	12($key),$s3
 	.byte	0xf3,0xc3			# rep ret
+.cfi_endproc
 .size	_x86_64_AES_encrypt_compact,.-_x86_64_AES_encrypt_compact
 ___
 
@@ -1161,6 +1163,7 @@ $code.=<<___;
 .type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
 .align	16
 _x86_64_AES_decrypt_compact:
+.cfi_startproc
 	lea	128($sbox),$inp			# size optimization
 	mov	0-128($inp),$acc1		# prefetch Td4
 	mov	32-128($inp),$acc2
@@ -1203,6 +1206,7 @@ $code.=<<___;
 	xor	8($key),$s2
 	xor	12($key),$s3
 	.byte	0xf3,0xc3			# rep ret
+.cfi_endproc
 .size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
 ___
 
@@ -1365,6 +1369,7 @@ AES_set_encrypt_key:
 .type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
 .align	16
 _x86_64_AES_set_encrypt_key:
+.cfi_startproc
 	mov	%esi,%ecx			# %ecx=bits
 	mov	%rdi,%rsi			# %rsi=userKey
 	mov	%rdx,%rdi			# %rdi=key
@@ -1546,6 +1551,7 @@ $code.=<<___;
 	mov	\$-1,%rax
 .Lexit:
 	.byte	0xf3,0xc3			# rep ret
+.cfi_endproc
 .size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
 ___
 
@@ -1728,7 +1734,9 @@ AES_cbc_encrypt:
 	cmp	\$0,%rdx	# check length
 	je	.Lcbc_epilogue
 	pushfq
-.cfi_push	49		# %rflags
+# This could be .cfi_push 49, but libunwind fails on registers it does not
+# recognize. See https://bugzilla.redhat.com/show_bug.cgi?id=217087.
+.cfi_adjust_cfa_offset	8
 	push	%rbx
 .cfi_push	%rbx
 	push	%rbp
@@ -1751,6 +1759,7 @@ AES_cbc_encrypt:
 	cmp	\$0,%r9
 	cmoveq	%r10,$sbox
 
+.cfi_remember_state
 	mov	OPENSSL_ia32cap_P(%rip),%r10d
 	cmp	\$$speed_limit,%rdx
 	jb	.Lcbc_slow_prologue
@@ -1986,6 +1995,7 @@ AES_cbc_encrypt:
 #--------------------------- SLOW ROUTINE ---------------------------#
 .align	16
 .Lcbc_slow_prologue:
+.cfi_restore_state
 	# allocate aligned stack frame...
 	lea	-88(%rsp),%rbp
 	and	\$-64,%rbp
@@ -1997,8 +2007,10 @@ AES_cbc_encrypt:
 	sub	%r10,%rbp
 
 	xchg	%rsp,%rbp
+.cfi_def_cfa_register	%rbp
 	#add	\$8,%rsp	# reserve for return address!
 	mov	%rbp,$_rsp	# save %rsp
+.cfi_cfa_expression	$_rsp,deref,+64
 .Lcbc_slow_body:
 	#mov	%rdi,$_inp	# save copy of inp
 	#mov	%rsi,$_out	# save copy of out
@@ -2187,7 +2199,9 @@ AES_cbc_encrypt:
 .cfi_def_cfa	%rsp,16
 .Lcbc_popfq:
 	popfq
-.cfi_pop	49		# %rflags
+# This could be .cfi_pop 49, but libunwind fails on registers it does not
+# recognize. See https://bugzilla.redhat.com/show_bug.cgi?id=217087.
+.cfi_adjust_cfa_offset	-8
 .Lcbc_epilogue:
 	ret
 .cfi_endproc
diff --git a/crypto/aes/asm/aesni-x86_64.pl b/crypto/aes/asm/aesni-x86_64.pl
index 2a202c53e5..ebcd9b5352 100644
--- a/crypto/aes/asm/aesni-x86_64.pl
+++ b/crypto/aes/asm/aesni-x86_64.pl
@@ -274,6 +274,7 @@ $code.=<<___;
 .type	${PREFIX}_encrypt,\@abi-omnipotent
 .align	16
 ${PREFIX}_encrypt:
+.cfi_startproc
 	movups	($inp),$inout0		# load input
 	mov	240($key),$rounds	# key->rounds
 ___
@@ -284,12 +285,14 @@ $code.=<<___;
 	movups	$inout0,($out)		# output
 	 pxor	$inout0,$inout0
 	ret
+.cfi_endproc
 .size	${PREFIX}_encrypt,.-${PREFIX}_encrypt
 
 .globl	${PREFIX}_decrypt
 .type	${PREFIX}_decrypt,\@abi-omnipotent
 .align	16
 ${PREFIX}_decrypt:
+.cfi_startproc
 	movups	($inp),$inout0		# load input
 	mov	240($key),$rounds	# key->rounds
 ___
@@ -300,6 +303,7 @@ $code.=<<___;
 	movups	$inout0,($out)		# output
 	 pxor	$inout0,$inout0
 	ret
+.cfi_endproc
 .size	${PREFIX}_decrypt, .-${PREFIX}_decrypt
 ___
 }
@@ -325,6 +329,7 @@ $code.=<<___;
 .type	_aesni_${dir}rypt2,\@abi-omnipotent
 .align	16
 _aesni_${dir}rypt2:
+.cfi_startproc
 	$movkey	($key),$rndkey0
 	shl	\$4,$rounds
 	$movkey	16($key),$rndkey1
@@ -350,6 +355,7 @@ _aesni_${dir}rypt2:
 	aes${dir}last	$rndkey0,$inout0
 	aes${dir}last	$rndkey0,$inout1
 	ret
+.cfi_endproc
 .size	_aesni_${dir}rypt2,.-_aesni_${dir}rypt2
 ___
 }
@@ -361,6 +367,7 @@ $code.=<<___;
 .type	_aesni_${dir}rypt3,\@abi-omnipotent
 .align	16
 _aesni_${dir}rypt3:
+.cfi_startproc
 	$movkey	($key),$rndkey0
 	shl	\$4,$rounds
 	$movkey	16($key),$rndkey1
@@ -391,6 +398,7 @@ _aesni_${dir}rypt3:
 	aes${dir}last	$rndkey0,$inout1
 	aes${dir}last	$rndkey0,$inout2
 	ret
+.cfi_endproc
 .size	_aesni_${dir}rypt3,.-_aesni_${dir}rypt3
 ___
 }
@@ -406,6 +414,7 @@ $code.=<<___;
 .type	_aesni_${dir}rypt4,\@abi-omnipotent
 .align	16
 _aesni_${dir}rypt4:
+.cfi_startproc
 	$movkey	($key),$rndkey0
 	shl	\$4,$rounds
 	$movkey	16($key),$rndkey1
@@ -442,6 +451,7 @@ _aesni_${dir}rypt4:
 	aes${dir}last	$rndkey0,$inout2
 	aes${dir}last	$rndkey0,$inout3
 	ret
+.cfi_endproc
 .size	_aesni_${dir}rypt4,.-_aesni_${dir}rypt4
 ___
 }
@@ -453,6 +463,7 @@ $code.=<<___;
 .type	_aesni_${dir}rypt6,\@abi-omnipotent
 .align	16
 _aesni_${dir}rypt6:
+.cfi_startproc
 	$movkey		($key),$rndkey0
 	shl		\$4,$rounds
 	$movkey		16($key),$rndkey1
@@ -503,6 +514,7 @@ _aesni_${dir}rypt6:
 	aes${dir}last	$rndkey0,$inout4
 	aes${dir}last	$rndkey0,$inout5
 	ret
+.cfi_endproc
 .size	_aesni_${dir}rypt6,.-_aesni_${dir}rypt6
 ___
 }
@@ -514,6 +526,7 @@ $code.=<<___;
 .type	_aesni_${dir}rypt8,\@abi-omnipotent
 .align	16
 _aesni_${dir}rypt8:
+.cfi_startproc
 	$movkey		($key),$rndkey0
 	shl		\$4,$rounds
 	$movkey		16($key),$rndkey1
@@ -574,6 +587,7 @@ _aesni_${dir}rypt8:
 	aes${dir}last	$rndkey0,$inout6
 	aes${dir}last	$rndkey0,$inout7
 	ret
+.cfi_endproc
 .size	_aesni_${dir}rypt8,.-_aesni_${dir}rypt8
 ___
 }
@@ -598,6 +612,7 @@ $code.=<<___;
 .type	aesni_ecb_encrypt,\@function,5
 .align	16
 aesni_ecb_encrypt:
+.cfi_startproc
 ___
 $code.=<<___ if ($win64);
 	lea	-0x58(%rsp),%rsp
@@ -943,6 +958,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	aesni_ecb_encrypt,.-aesni_ecb_encrypt
 ___
 
diff --git a/crypto/aes/asm/bsaes-x86_64.pl b/crypto/aes/asm/bsaes-x86_64.pl
index 2c79c2b67c..030452b1fd 100644
--- a/crypto/aes/asm/bsaes-x86_64.pl
+++ b/crypto/aes/asm/bsaes-x86_64.pl
@@ -816,6 +816,7 @@ $code.=<<___;
 .type	_bsaes_encrypt8,\@abi-omnipotent
 .align	64
 _bsaes_encrypt8:
+.cfi_startproc
 	lea	.LBS0(%rip), $const	# constants table
 
 	movdqa	($key), @XMM[9]		# round 0 key
@@ -875,11 +876,13 @@ $code.=<<___;
 	pxor	@XMM[8], @XMM[0]
 	pxor	@XMM[8], @XMM[1]
 	ret
+.cfi_endproc
 .size	_bsaes_encrypt8,.-_bsaes_encrypt8
 
 .type	_bsaes_decrypt8,\@abi-omnipotent
 .align	64
 _bsaes_decrypt8:
+.cfi_startproc
 	lea	.LBS0(%rip), $const	# constants table
 
 	movdqa	($key), @XMM[9]		# round 0 key
@@ -937,6 +940,7 @@ $code.=<<___;
 	pxor	@XMM[8], @XMM[0]
 	pxor	@XMM[8], @XMM[1]
 	ret
+.cfi_endproc
 .size	_bsaes_decrypt8,.-_bsaes_decrypt8
 ___
 }
@@ -971,6 +975,7 @@ $code.=<<___;
 .type	_bsaes_key_convert,\@abi-omnipotent
 .align	16
 _bsaes_key_convert:
+.cfi_startproc
 	lea	.Lmasks(%rip), $const
 	movdqu	($inp), %xmm7		# load round 0 key
 	lea	0x10($inp), $inp
@@ -1049,6 +1054,7 @@ _bsaes_key_convert:
 	movdqa	0x50($const), %xmm7	# .L63
 	#movdqa	%xmm6, ($out)		# don't save last round key
 	ret
+.cfi_endproc
 .size	_bsaes_key_convert,.-_bsaes_key_convert
 ___
 }
diff --git a/crypto/aes/asm/vpaes-x86_64.pl b/crypto/aes/asm/vpaes-x86_64.pl
index b715aca167..ec2a8da819 100644
--- a/crypto/aes/asm/vpaes-x86_64.pl
+++ b/crypto/aes/asm/vpaes-x86_64.pl
@@ -91,6 +91,7 @@ $code.=<<___;
 .type	_vpaes_encrypt_core,\@abi-omnipotent
 .align 16
 _vpaes_encrypt_core:
+.cfi_startproc
 	mov	%rdx,	%r9
 	mov	\$16,	%r11
 	mov	240(%rdx),%eax
@@ -171,6 +172,7 @@ _vpaes_encrypt_core:
 	pxor	%xmm4,	%xmm0	# 0 = A
 	pshufb	%xmm1,	%xmm0
 	ret
+.cfi_endproc
 .size	_vpaes_encrypt_core,.-_vpaes_encrypt_core
 
 ##
@@ -181,6 +183,7 @@ _vpaes_encrypt_core:
 .type	_vpaes_decrypt_core,\@abi-omnipotent
 .align	16
 _vpaes_decrypt_core:
+.cfi_startproc
 	mov	%rdx,	%r9		# load key
 	mov	240(%rdx),%eax
 	movdqa	%xmm9,	%xmm1
@@ -277,6 +280,7 @@ _vpaes_decrypt_core:
 	pxor	%xmm4,	%xmm0	# 0 = A
 	pshufb	%xmm2,	%xmm0
 	ret
+.cfi_endproc
 .size	_vpaes_decrypt_core,.-_vpaes_decrypt_core
 
 ########################################################
@@ -287,6 +291,7 @@ _vpaes_decrypt_core:
 .type	_vpaes_schedule_core,\@abi-omnipotent
 .align	16
 _vpaes_schedule_core:
+.cfi_startproc
 	# rdi = key
 	# rsi = size in bits
 	# rdx = buffer
@@ -453,6 +458,7 @@ _vpaes_schedule_core:
 	pxor	%xmm6,  %xmm6
 	pxor	%xmm7,  %xmm7
 	ret
+.cfi_endproc
 .size	_vpaes_schedule_core,.-_vpaes_schedule_core
 
 ##
@@ -472,6 +478,7 @@ _vpaes_schedule_core:
 .type	_vpaes_schedule_192_smear,\@abi-omnipotent
 .align	16
 _vpaes_schedule_192_smear:
+.cfi_startproc
 	pshufd	\$0x80,	%xmm6,	%xmm1	# d c 0 0 -> c 0 0 0
 	pshufd	\$0xFE,	%xmm7,	%xmm0	# b a _ _ -> b b b a
 	pxor	%xmm1,	%xmm6		# -> c+d c 0 0
@@ -480,6 +487,7 @@ _vpaes_schedule_192_smear:
 	movdqa	%xmm6,	%xmm0
 	movhlps	%xmm1,	%xmm6		# clobber low side with zeros
 	ret
+.cfi_endproc
 .size	_vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear
 
 ##
@@ -503,6 +511,7 @@ _vpaes_schedule_192_smear:
 .type	_vpaes_schedule_round,\@abi-omnipotent
 .align	16
 _vpaes_schedule_round:
+.cfi_startproc
 	# extract rcon from xmm8
 	pxor	%xmm1,	%xmm1
 	palignr	\$15,	%xmm8,	%xmm1
@@ -556,6 +565,7 @@ _vpaes_schedule_low_round:
 	pxor	%xmm7,	%xmm0
 	movdqa	%xmm0,	%xmm7
 	ret
+.cfi_endproc
 .size	_vpaes_schedule_round,.-_vpaes_schedule_round
 
 ##
@@ -570,6 +580,7 @@ _vpaes_schedule_low_round:
 .type	_vpaes_schedule_transform,\@abi-omnipotent
 .align	16
 _vpaes_schedule_transform:
+.cfi_startproc
 	movdqa	%xmm9,	%xmm1
 	pandn	%xmm0,	%xmm1
 	psrld	\$4,	%xmm1
@@ -580,6 +591,7 @@ _vpaes_schedule_transform:
 	pshufb	%xmm1,	%xmm0
 	pxor	%xmm2,	%xmm0
 	ret
+.cfi_endproc
 .size	_vpaes_schedule_transform,.-_vpaes_schedule_transform
 
 ##
@@ -608,6 +620,7 @@ _vpaes_schedule_transform:
 .type	_vpaes_schedule_mangle,\@abi-omnipotent
 .align	16
 _vpaes_schedule_mangle:
+.cfi_startproc
 	movdqa	%xmm0,	%xmm4	# save xmm0 for later
 	movdqa	.Lk_mc_forward(%rip),%xmm5
 	test	%rcx, 	%rcx
@@ -672,6 +685,7 @@ _vpaes_schedule_mangle:
 	and	\$0x30,	%r8
 	movdqu	%xmm3,	(%rdx)
 	ret
+.cfi_endproc
 .size	_vpaes_schedule_mangle,.-_vpaes_schedule_mangle
 
 #
@@ -681,6 +695,7 @@ _vpaes_schedule_mangle:
 .type	${PREFIX}_set_encrypt_key,\@function,3
 .align	16
 ${PREFIX}_set_encrypt_key:
+.cfi_startproc
 ___
 $code.=<<___ if ($win64);
 	lea	-0xb8(%rsp),%rsp
@@ -723,12 +738,14 @@ ___
 $code.=<<___;
 	xor	%eax,%eax
 	ret
+.cfi_endproc
 .size	${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
 
 .globl	${PREFIX}_set_decrypt_key
 .type	${PREFIX}_set_decrypt_key,\@function,3
 .align	16
 ${PREFIX}_set_decrypt_key:
+.cfi_startproc
 ___
 $code.=<<___ if ($win64);
 	lea	-0xb8(%rsp),%rsp
@@ -776,12 +793,14 @@ ___
 $code.=<<___;
 	xor	%eax,%eax
 	ret
+.cfi_endproc
 .size	${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
 
 .globl	${PREFIX}_encrypt
 .type	${PREFIX}_encrypt,\@function,3
 .align	16
 ${PREFIX}_encrypt:
+.cfi_startproc
 ___
 $code.=<<___ if ($win64);
 	lea	-0xb8(%rsp),%rsp
@@ -819,12 +838,14 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	${PREFIX}_encrypt,.-${PREFIX}_encrypt
 
 .globl	${PREFIX}_decrypt
 .type	${PREFIX}_decrypt,\@function,3
 .align	16
 ${PREFIX}_decrypt:
+.cfi_startproc
 ___
 $code.=<<___ if ($win64);
 	lea	-0xb8(%rsp),%rsp
@@ -862,6 +883,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	${PREFIX}_decrypt,.-${PREFIX}_decrypt
 ___
 {
@@ -874,6 +896,7 @@ $code.=<<___;
 .type	${PREFIX}_cbc_encrypt,\@function,6
 .align	16
 ${PREFIX}_cbc_encrypt:
+.cfi_startproc
 	xchg	$key,$len
 ___
 ($len,$key)=($key,$len);
@@ -944,6 +967,7 @@ ___
 $code.=<<___;
 .Lcbc_abort:
 	ret
+.cfi_endproc
 .size	${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
 ___
 }
@@ -957,6 +981,7 @@ $code.=<<___;
 .type	_vpaes_preheat,\@abi-omnipotent
 .align	16
 _vpaes_preheat:
+.cfi_startproc
 	lea	.Lk_s0F(%rip), %r10
 	movdqa	-0x20(%r10), %xmm10	# .Lk_inv
 	movdqa	-0x10(%r10), %xmm11	# .Lk_inv+16
@@ -966,6 +991,7 @@ _vpaes_preheat:
 	movdqa	0x50(%r10), %xmm15	# .Lk_sb2
 	movdqa	0x60(%r10), %xmm14	# .Lk_sb2+16
 	ret
+.cfi_endproc
 .size	_vpaes_preheat,.-_vpaes_preheat
 ########################################################
 ##                                                    ##
diff --git a/crypto/bn/asm/rsaz-avx2.pl b/crypto/bn/asm/rsaz-avx2.pl
index f1292cc75c..b2958715e6 100755
--- a/crypto/bn/asm/rsaz-avx2.pl
+++ b/crypto/bn/asm/rsaz-avx2.pl
@@ -1492,6 +1492,7 @@ $code.=<<___;
 .type	rsaz_1024_red2norm_avx2,\@abi-omnipotent
 .align	32
 rsaz_1024_red2norm_avx2:
+.cfi_startproc
 	sub	\$-128,$inp	# size optimization
 	xor	%rax,%rax
 ___
@@ -1525,12 +1526,14 @@ ___
 }
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	rsaz_1024_red2norm_avx2,.-rsaz_1024_red2norm_avx2
 
 .globl	rsaz_1024_norm2red_avx2
 .type	rsaz_1024_norm2red_avx2,\@abi-omnipotent
 .align	32
 rsaz_1024_norm2red_avx2:
+.cfi_startproc
 	sub	\$-128,$out	# size optimization
 	mov	($inp),@T[0]
 	mov	\$0x1fffffff,%eax
@@ -1562,6 +1565,7 @@ $code.=<<___;
 	mov	@T[0],`8*($j+2)-128`($out)
 	mov	@T[0],`8*($j+3)-128`($out)
 	ret
+.cfi_endproc
 .size	rsaz_1024_norm2red_avx2,.-rsaz_1024_norm2red_avx2
 ___
 }
@@ -1573,6 +1577,7 @@ $code.=<<___;
 .type	rsaz_1024_scatter5_avx2,\@abi-omnipotent
 .align	32
 rsaz_1024_scatter5_avx2:
+.cfi_startproc
 	vzeroupper
 	vmovdqu	.Lscatter_permd(%rip),%ymm5
 	shl	\$4,$power
@@ -1592,6 +1597,7 @@ rsaz_1024_scatter5_avx2:
 
 	vzeroupper
 	ret
+.cfi_endproc
 .size	rsaz_1024_scatter5_avx2,.-rsaz_1024_scatter5_avx2
 
 .globl	rsaz_1024_gather5_avx2
diff --git a/crypto/bn/asm/x86_64-mont5.pl b/crypto/bn/asm/x86_64-mont5.pl
index ad6e8ada3c..0492690491 100755
--- a/crypto/bn/asm/x86_64-mont5.pl
+++ b/crypto/bn/asm/x86_64-mont5.pl
@@ -2910,6 +2910,7 @@ bn_powerx5:
 .align	32
 bn_sqrx8x_internal:
 __bn_sqrx8x_internal:
+.cfi_startproc
 	##################################################################
 	# Squaring part:
 	#
@@ -3542,6 +3543,7 @@ __bn_sqrx8x_reduction:
 	cmp	8+8(%rsp),%r8		# end of t[]?
 	jb	.Lsqrx8x_reduction_loop
 	ret
+.cfi_endproc
 .size	bn_sqrx8x_internal,.-bn_sqrx8x_internal
 ___
 }
diff --git a/crypto/ec/asm/ecp_nistz256-x86_64.pl b/crypto/ec/asm/ecp_nistz256-x86_64.pl
index eba6ffd430..b1ff2e38a7 100755
--- a/crypto/ec/asm/ecp_nistz256-x86_64.pl
+++ b/crypto/ec/asm/ecp_nistz256-x86_64.pl
@@ -1674,6 +1674,7 @@ $code.=<<___;
 .type	__ecp_nistz256_mul_montq,\@abi-omnipotent
 .align	32
 __ecp_nistz256_mul_montq:
+.cfi_startproc
 	########################################################################
 	# Multiply a by b[0]
 	mov	%rax, $t1
@@ -1885,6 +1886,7 @@ __ecp_nistz256_mul_montq:
 	mov	$acc1, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_mul_montq,.-__ecp_nistz256_mul_montq
 
 ################################################################################
@@ -1968,6 +1970,7 @@ $code.=<<___;
 .type	__ecp_nistz256_sqr_montq,\@abi-omnipotent
 .align	32
 __ecp_nistz256_sqr_montq:
+.cfi_startproc
 	mov	%rax, $acc5
 	mulq	$acc6			# a[1]*a[0]
 	mov	%rax, $acc1
@@ -2125,6 +2128,7 @@ __ecp_nistz256_sqr_montq:
 	mov	$acc7, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_sqr_montq,.-__ecp_nistz256_sqr_montq
 ___
 
@@ -2133,6 +2137,7 @@ $code.=<<___;
 .type	__ecp_nistz256_mul_montx,\@abi-omnipotent
 .align	32
 __ecp_nistz256_mul_montx:
+.cfi_startproc
 	########################################################################
 	# Multiply by b[0]
 	mulx	$acc1, $acc0, $acc1
@@ -2295,11 +2300,13 @@ __ecp_nistz256_mul_montx:
 	mov	$acc1, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_mul_montx,.-__ecp_nistz256_mul_montx
 
 .type	__ecp_nistz256_sqr_montx,\@abi-omnipotent
 .align	32
 __ecp_nistz256_sqr_montx:
+.cfi_startproc
 	mulx	$acc6, $acc1, $acc2	# a[0]*a[1]
 	mulx	$acc7, $t0, $acc3	# a[0]*a[2]
 	xor	%eax, %eax
@@ -2423,6 +2430,7 @@ __ecp_nistz256_sqr_montx:
 	mov	$acc7, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_sqr_montx,.-__ecp_nistz256_sqr_montx
 ___
 }
@@ -2578,6 +2586,7 @@ ecp_nistz256_scatter_w5:
 .type	ecp_nistz256_gather_w5,\@abi-omnipotent
 .align	32
 ecp_nistz256_gather_w5:
+.cfi_startproc
 ___
 $code.=<<___	if ($avx>1);
 	mov	OPENSSL_ia32cap_P+8(%rip), %eax
@@ -2666,6 +2675,7 @@ $code.=<<___	if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .LSEH_end_ecp_nistz256_gather_w5:
 .size	ecp_nistz256_gather_w5,.-ecp_nistz256_gather_w5
 
@@ -2694,6 +2704,7 @@ ecp_nistz256_scatter_w7:
 .type	ecp_nistz256_gather_w7,\@abi-omnipotent
 .align	32
 ecp_nistz256_gather_w7:
+.cfi_startproc
 ___
 $code.=<<___	if ($avx>1);
 	mov	OPENSSL_ia32cap_P+8(%rip), %eax
@@ -2771,6 +2782,7 @@ $code.=<<___	if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .LSEH_end_ecp_nistz256_gather_w7:
 .size	ecp_nistz256_gather_w7,.-ecp_nistz256_gather_w7
 ___
@@ -2787,6 +2799,7 @@ $code.=<<___;
 .type	ecp_nistz256_avx2_gather_w5,\@abi-omnipotent
 .align	32
 ecp_nistz256_avx2_gather_w5:
+.cfi_startproc
 .Lavx2_gather_w5:
 	vzeroupper
 ___
@@ -2874,6 +2887,7 @@ $code.=<<___	if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .LSEH_end_ecp_nistz256_avx2_gather_w5:
 .size	ecp_nistz256_avx2_gather_w5,.-ecp_nistz256_avx2_gather_w5
 ___
@@ -2893,6 +2907,7 @@ $code.=<<___;
 .type	ecp_nistz256_avx2_gather_w7,\@abi-omnipotent
 .align	32
 ecp_nistz256_avx2_gather_w7:
+.cfi_startproc
 .Lavx2_gather_w7:
 	vzeroupper
 ___
@@ -2995,6 +3010,7 @@ $code.=<<___	if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .LSEH_end_ecp_nistz256_avx2_gather_w7:
 .size	ecp_nistz256_avx2_gather_w7,.-ecp_nistz256_avx2_gather_w7
 ___
@@ -3064,6 +3080,7 @@ $code.=<<___;
 .type	__ecp_nistz256_add_toq,\@abi-omnipotent
 .align	32
 __ecp_nistz256_add_toq:
+.cfi_startproc
 	xor	$t4,$t4
 	add	8*0($b_ptr), $a0
 	adc	8*1($b_ptr), $a1
@@ -3091,11 +3108,13 @@ __ecp_nistz256_add_toq:
 	mov	$a3, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_add_toq,.-__ecp_nistz256_add_toq
 
 .type	__ecp_nistz256_sub_fromq,\@abi-omnipotent
 .align	32
 __ecp_nistz256_sub_fromq:
+.cfi_startproc
 	sub	8*0($b_ptr), $a0
 	sbb	8*1($b_ptr), $a1
 	 mov	$a0, $t0
@@ -3122,11 +3141,13 @@ __ecp_nistz256_sub_fromq:
 	mov	$a3, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_sub_fromq,.-__ecp_nistz256_sub_fromq
 
 .type	__ecp_nistz256_subq,\@abi-omnipotent
 .align	32
 __ecp_nistz256_subq:
+.cfi_startproc
 	sub	$a0, $t0
 	sbb	$a1, $t1
 	 mov	$t0, $a0
@@ -3149,11 +3170,13 @@ __ecp_nistz256_subq:
 	cmovnz	$t3, $a3
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_subq,.-__ecp_nistz256_subq
 
 .type	__ecp_nistz256_mul_by_2q,\@abi-omnipotent
 .align	32
 __ecp_nistz256_mul_by_2q:
+.cfi_startproc
 	xor	$t4, $t4
 	add	$a0, $a0		# a0:a3+a0:a3
 	adc	$a1, $a1
@@ -3181,6 +3204,7 @@ __ecp_nistz256_mul_by_2q:
 	mov	$a3, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_mul_by_2q,.-__ecp_nistz256_mul_by_2q
 ___
 									}
@@ -3620,7 +3644,9 @@ $code.=<<___;
 	movq	%xmm1, $a_ptr			# restore $a_ptr
 	movq	%xmm0, $r_ptr			# restore $r_ptr
 	add	\$`32*(18-5)`, %rsp		# difference in frame sizes
+.cfi_adjust_cfa_offset	`-32*(18-5)`
 	jmp	.Lpoint_double_shortcut$x
+.cfi_adjust_cfa_offset	`32*(18-5)`
 
 .align	32
 .Ladd_proceed$x:
@@ -4156,6 +4182,7 @@ $code.=<<___;
 .type	__ecp_nistz256_add_tox,\@abi-omnipotent
 .align	32
 __ecp_nistz256_add_tox:
+.cfi_startproc
 	xor	$t4, $t4
 	adc	8*0($b_ptr), $a0
 	adc	8*1($b_ptr), $a1
@@ -4184,11 +4211,13 @@ __ecp_nistz256_add_tox:
 	mov	$a3, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_add_tox,.-__ecp_nistz256_add_tox
 
 .type	__ecp_nistz256_sub_fromx,\@abi-omnipotent
 .align	32
 __ecp_nistz256_sub_fromx:
+.cfi_startproc
 	xor	$t4, $t4
 	sbb	8*0($b_ptr), $a0
 	sbb	8*1($b_ptr), $a1
@@ -4217,11 +4246,13 @@ __ecp_nistz256_sub_fromx:
 	mov	$a3, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_sub_fromx,.-__ecp_nistz256_sub_fromx
 
 .type	__ecp_nistz256_subx,\@abi-omnipotent
 .align	32
 __ecp_nistz256_subx:
+.cfi_startproc
 	xor	$t4, $t4
 	sbb	$a0, $t0
 	sbb	$a1, $t1
@@ -4246,11 +4277,13 @@ __ecp_nistz256_subx:
 	cmovc	$t3, $a3
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_subx,.-__ecp_nistz256_subx
 
 .type	__ecp_nistz256_mul_by_2x,\@abi-omnipotent
 .align	32
 __ecp_nistz256_mul_by_2x:
+.cfi_startproc
 	xor	$t4, $t4
 	adc	$a0, $a0		# a0:a3+a0:a3
 	adc	$a1, $a1
@@ -4279,6 +4312,7 @@ __ecp_nistz256_mul_by_2x:
 	mov	$a3, 8*3($r_ptr)
 
 	ret
+.cfi_endproc
 .size	__ecp_nistz256_mul_by_2x,.-__ecp_nistz256_mul_by_2x
 ___
 									}
diff --git a/crypto/modes/asm/ghash-x86_64.pl b/crypto/modes/asm/ghash-x86_64.pl
index afc30c3e72..136b5a394e 100644
--- a/crypto/modes/asm/ghash-x86_64.pl
+++ b/crypto/modes/asm/ghash-x86_64.pl
@@ -529,6 +529,7 @@ $code.=<<___;
 .type	gcm_init_clmul,\@abi-omnipotent
 .align	16
 gcm_init_clmul:
+.cfi_startproc
 .L_init_clmul:
 ___
 $code.=<<___ if ($win64);
@@ -598,6 +599,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	gcm_init_clmul,.-gcm_init_clmul
 ___
 }
@@ -609,6 +611,7 @@ $code.=<<___;
 .type	gcm_gmult_clmul,\@abi-omnipotent
 .align	16
 gcm_gmult_clmul:
+.cfi_startproc
 .L_gmult_clmul:
 	movdqu		($Xip),$Xi
 	movdqa		.Lbswap_mask(%rip),$T3
@@ -645,6 +648,7 @@ $code.=<<___;
 	pshufb		$T3,$Xi
 	movdqu		$Xi,($Xip)
 	ret
+.cfi_endproc
 .size	gcm_gmult_clmul,.-gcm_gmult_clmul
 ___
 }
@@ -658,6 +662,7 @@ $code.=<<___;
 .type	gcm_ghash_clmul,\@abi-omnipotent
 .align	32
 gcm_ghash_clmul:
+.cfi_startproc
 .L_ghash_clmul:
 ___
 $code.=<<___ if ($win64);
@@ -1005,6 +1010,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	gcm_ghash_clmul,.-gcm_ghash_clmul
 ___
 }
@@ -1014,6 +1020,7 @@ $code.=<<___;
 .type	gcm_init_avx,\@abi-omnipotent
 .align	32
 gcm_init_avx:
+.cfi_startproc
 ___
 if ($avx) {
 my ($Htbl,$Xip)=@_4args;
@@ -1142,6 +1149,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	gcm_init_avx,.-gcm_init_avx
 ___
 } else {
@@ -1156,7 +1164,9 @@ $code.=<<___;
 .type	gcm_gmult_avx,\@abi-omnipotent
 .align	32
 gcm_gmult_avx:
+.cfi_startproc
 	jmp	.L_gmult_clmul
+.cfi_endproc
 .size	gcm_gmult_avx,.-gcm_gmult_avx
 ___
 
@@ -1165,6 +1175,7 @@ $code.=<<___;
 .type	gcm_ghash_avx,\@abi-omnipotent
 .align	32
 gcm_ghash_avx:
+.cfi_startproc
 ___
 if ($avx) {
 my ($Xip,$Htbl,$inp,$len)=@_4args;
@@ -1577,6 +1588,7 @@ $code.=<<___ if ($win64);
 ___
 $code.=<<___;
 	ret
+.cfi_endproc
 .size	gcm_ghash_avx,.-gcm_ghash_avx
 ___
 } else {
diff --git a/crypto/perlasm/x86_64-xlate.pl b/crypto/perlasm/x86_64-xlate.pl
index f8380f2e9c..b81d03a753 100755
--- a/crypto/perlasm/x86_64-xlate.pl
+++ b/crypto/perlasm/x86_64-xlate.pl
@@ -541,6 +541,7 @@ my %globals;
 	);
 
     my ($cfa_reg, $cfa_rsp);
+    my @cfa_stack;
 
     # [us]leb128 format is variable-length integer representation base
     # 2^128, with most significant bit of each byte being 0 denoting
@@ -688,6 +689,14 @@ my %globals;
 						      cfa_expression($$line)));
 				last;
 			      };
+	    /remember_state/
+			&& do {	push @cfa_stack, [$cfa_reg, $cfa_rsp];
+				last;
+			      };
+	    /restore_state/
+			&& do {	($cfa_reg, $cfa_rsp) = @{pop @cfa_stack};
+				last;
+			      };
 	    }
 
 	    $self->{value} = ".cfi_$dir\t$$line" if ($dir);