Add CFI unwind assembly directives for 64-bit ARM assembly

* cipher/asm-common-aarch64.h (CFI_STARTPROC, CFI_ENDPROC)
(CFI_REMEMBER_STATE, CFI_RESTORE_STATE, CFI_ADJUST_CFA_OFFSET)
(CFI_REL_OFFSET, CFI_DEF_CFA_REGISTER, CFI_REGISTER, CFI_RESTORE)
(DW_REGNO_SP, DW_SLEB128_7BIT, DW_SLEB128_28BIT, CFI_CFA_ON_STACK)
(CFI_REG_ON_STACK): New.
* cipher/camellia-aarch64.S: Add CFI directives.
* cipher/chacha20-aarch64.S: Add CFI directives.
* cipher/cipher-gcm-armv8-aarch64-ce.S: Add CFI directives.
* cipher/crc-armv8-aarch64-ce.S: Add CFI directives.
* cipher/rijndael-aarch64.S: Add CFI directives.
* cipher/rijndael-armv8-aarch64-ce.S: Add CFI directives.
* cipher/sha1-armv8-aarch64-ce.S: Add CFI directives.
* cipher/sha256-armv8-aarch64-ce.S: Add CFI directives.
* cipher/twofish-aarch64.S: Add CFI directives.
* mpi/aarch64/mpih-add1.S: Add CFI directives.
* mpi/aarch64/mpih-mul1.S: Add CFI directives.
* mpi/aarch64/mpih-mul2.S: Add CFI directives.
* mpi/aarch64/mpih-mul3.S: Add CFI directives.
* mpi/aarch64/mpih-sub1.S: Add CFI directives.
* mpi/asm-common-aarch64.h: Include "../cipher/asm-common-aarch64.h".
(ELF): Remove.
--

This commit adds CFI directives that add DWARF unwinding information for
debugger to backtrace when executing code from 64-bit ARM assembly files.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

16 files changed, 160 insertions, 10 deletions

diff --git a/cipher/asm-common-aarch64.h b/cipher/asm-common-aarch64.h
index 814b7ad1..502c35ae 100644
--- a/cipher/asm-common-aarch64.h
+++ b/cipher/asm-common-aarch64.h
@@ -29,4 +29,62 @@
 # define ELF(...) /*_*/
 #endif
 
+#ifdef HAVE_GCC_ASM_CFI_DIRECTIVES
+/* CFI directives to emit DWARF stack unwinding information. */
+# define CFI_STARTPROC()            .cfi_startproc
+# define CFI_ENDPROC()              .cfi_endproc
+# define CFI_REMEMBER_STATE()       .cfi_remember_state
+# define CFI_RESTORE_STATE()        .cfi_restore_state
+# define CFI_ADJUST_CFA_OFFSET(off) .cfi_adjust_cfa_offset off
+# define CFI_REL_OFFSET(reg,off)    .cfi_rel_offset reg, off
+# define CFI_DEF_CFA_REGISTER(reg)  .cfi_def_cfa_register reg
+# define CFI_REGISTER(ro,rn)        .cfi_register ro, rn
+# define CFI_RESTORE(reg)           .cfi_restore reg
+
+/* CFA expressions are used for pointing CFA and registers to
+ * SP relative offsets. */
+# define DW_REGNO_SP 31
+
+/* Fixed length encoding used for integers for now. */
+# define DW_SLEB128_7BIT(value) \
+	0x00|((value) & 0x7f)
+# define DW_SLEB128_28BIT(value) \
+	0x80|((value)&0x7f), \
+	0x80|(((value)>>7)&0x7f), \
+	0x80|(((value)>>14)&0x7f), \
+	0x00|(((value)>>21)&0x7f)
+
+# define CFI_CFA_ON_STACK(rsp_offs,cfa_depth) \
+	.cfi_escape \
+	  0x0f, /* DW_CFA_def_cfa_expression */ \
+	    DW_SLEB128_7BIT(11), /* length */ \
+	  0x8f, /* DW_OP_breg31, rsp + constant */ \
+	    DW_SLEB128_28BIT(rsp_offs), \
+	  0x06, /* DW_OP_deref */ \
+	  0x23, /* DW_OP_plus_constu */ \
+	    DW_SLEB128_28BIT((cfa_depth)+8)
+
+# define CFI_REG_ON_STACK(regno,rsp_offs) \
+	.cfi_escape \
+	  0x10, /* DW_CFA_expression */ \
+	    DW_SLEB128_7BIT(regno), \
+	    DW_SLEB128_7BIT(5), /* length */ \
+	  0x8f, /* DW_OP_breg31, rsp + constant */ \
+	    DW_SLEB128_28BIT(rsp_offs)
+
+#else
+# define CFI_STARTPROC()
+# define CFI_ENDPROC()
+# define CFI_REMEMBER_STATE()
+# define CFI_RESTORE_STATE()
+# define CFI_ADJUST_CFA_OFFSET(off)
+# define CFI_REL_OFFSET(reg,off)
+# define CFI_DEF_CFA_REGISTER(reg)
+# define CFI_REGISTER(ro,rn)
+# define CFI_RESTORE(reg)
+
+# define CFI_CFA_ON_STACK(rsp_offs,cfa_depth)
+# define CFI_REG_ON_STACK(reg,rsp_offs)
+#endif
+
 #endif /* GCRY_ASM_COMMON_AARCH64_H */
diff --git a/cipher/camellia-aarch64.S b/cipher/camellia-aarch64.S
index 5c6ab020..f4980862 100644
--- a/cipher/camellia-aarch64.S
+++ b/cipher/camellia-aarch64.S
@@ -201,7 +201,12 @@
 ELF(.type   _gcry_camellia_arm_encrypt_block,@function;)
 
 _gcry_camellia_arm_encrypt_block:
+	CFI_STARTPROC()
 	stp x19, x30, [sp, #-16]!
+	CFI_ADJUST_CFA_OFFSET(16)
+	CFI_REG_ON_STACK(19, 0)
+	CFI_REG_ON_STACK(30, 8)
+
 	/* input:
 	 *	x0: keytable
 	 *	x1: dst
@@ -228,8 +233,13 @@ _gcry_camellia_arm_encrypt_block:
 
 	outunpack(24);
 
+	CFI_REMEMBER_STATE()
 	ldp x19, x30, [sp], #16
+	CFI_ADJUST_CFA_OFFSET(-16)
+	CFI_RESTORE(x19)
+	CFI_RESTORE(x30)
 	ret;
+	CFI_RESTORE_STATE()
 .ltorg
 
 .Lenc_256:
@@ -239,7 +249,11 @@ _gcry_camellia_arm_encrypt_block:
 	outunpack(32);
 
 	ldp x19, x30, [sp], #16
+	CFI_ADJUST_CFA_OFFSET(-16)
+	CFI_RESTORE(x19)
+	CFI_RESTORE(x30)
 	ret;
+	CFI_ENDPROC()
 .ltorg
 ELF(.size _gcry_camellia_arm_encrypt_block,.-_gcry_camellia_arm_encrypt_block;)
 
@@ -247,7 +261,12 @@ ELF(.size _gcry_camellia_arm_encrypt_block,.-_gcry_camellia_arm_encrypt_block;)
 ELF(.type   _gcry_camellia_arm_decrypt_block,@function;)
 
 _gcry_camellia_arm_decrypt_block:
+	CFI_STARTPROC()
 	stp x19, x30, [sp, #-16]!
+	CFI_ADJUST_CFA_OFFSET(16)
+	CFI_REG_ON_STACK(19, 0)
+	CFI_REG_ON_STACK(30, 8)
+
 	/* input:
 	 *	x0: keytable
 	 *	x1: dst
@@ -275,8 +294,13 @@ _gcry_camellia_arm_decrypt_block:
 
 	outunpack(0);
 
+	CFI_REMEMBER_STATE()
 	ldp x19, x30, [sp], #16
+	CFI_ADJUST_CFA_OFFSET(-16)
+	CFI_RESTORE(x19)
+	CFI_RESTORE(x30)
 	ret;
+	CFI_RESTORE_STATE()
 .ltorg
 
 .Ldec_256:
@@ -285,6 +309,7 @@ _gcry_camellia_arm_decrypt_block:
 	dec_fls(24);
 
 	b .Ldec_128;
+	CFI_ENDPROC()
 .ltorg
 ELF(.size _gcry_camellia_arm_decrypt_block,.-_gcry_camellia_arm_decrypt_block;)
 
diff --git a/cipher/chacha20-aarch64.S b/cipher/chacha20-aarch64.S
index 3844d4e1..adb9b1f2 100644
--- a/cipher/chacha20-aarch64.S
+++ b/cipher/chacha20-aarch64.S
@@ -163,6 +163,7 @@ _gcry_chacha20_aarch64_blocks4:
 	 *	x2: src
 	 *	x3: nblks (multiple of 4)
 	 */
+	CFI_STARTPROC()
 
 	GET_DATA_POINTER(CTR, .Linc_counter);
 	add INPUT_CTR, INPUT, #(12*4);
@@ -309,6 +310,7 @@ _gcry_chacha20_aarch64_blocks4:
 
 	eor x0, x0, x0
 	ret
+	CFI_ENDPROC()
 ELF(.size _gcry_chacha20_aarch64_blocks4, .-_gcry_chacha20_aarch64_blocks4;)
 
 #endif
diff --git a/cipher/cipher-gcm-armv8-aarch64-ce.S b/cipher/cipher-gcm-armv8-aarch64-ce.S
index b6c4f59d..7c6be94e 100644
--- a/cipher/cipher-gcm-armv8-aarch64-ce.S
+++ b/cipher/cipher-gcm-armv8-aarch64-ce.S
@@ -157,15 +157,23 @@ gcry_gcm_reduction_constant:
 
 #define VPUSH_ABI \
         stp d8, d9, [sp, #-16]!; \
+        CFI_ADJUST_CFA_OFFSET(16); \
         stp d10, d11, [sp, #-16]!; \
+        CFI_ADJUST_CFA_OFFSET(16); \
         stp d12, d13, [sp, #-16]!; \
-        stp d14, d15, [sp, #-16]!;
+        CFI_ADJUST_CFA_OFFSET(16); \
+        stp d14, d15, [sp, #-16]!; \
+        CFI_ADJUST_CFA_OFFSET(16);
 
 #define VPOP_ABI \
         ldp d14, d15, [sp], #16; \
+        CFI_ADJUST_CFA_OFFSET(-16); \
         ldp d12, d13, [sp], #16; \
+        CFI_ADJUST_CFA_OFFSET(-16); \
         ldp d10, d11, [sp], #16; \
-        ldp d8, d9, [sp], #16;
+        CFI_ADJUST_CFA_OFFSET(-16); \
+        ldp d8, d9, [sp], #16; \
+        CFI_ADJUST_CFA_OFFSET(-16);
 
 /*
  * unsigned int _gcry_ghash_armv8_ce_pmull (void *gcm_key, byte *result,
@@ -183,6 +191,8 @@ _gcry_ghash_armv8_ce_pmull:
    *    x3: nblocks
    *    x4: gcm_table
    */
+  CFI_STARTPROC();
+
   cbz x3, .Ldo_nothing;
 
   GET_DATA_POINTER(x5, .Lrconst)
@@ -360,6 +370,7 @@ _gcry_ghash_armv8_ce_pmull:
 .Ldo_nothing:
   mov x0, #0
   ret
+  CFI_ENDPROC()
 ELF(.size _gcry_ghash_armv8_ce_pmull,.-_gcry_ghash_armv8_ce_pmull;)
 
 
@@ -374,6 +385,7 @@ _gcry_ghash_setup_armv8_ce_pmull:
    *	x0: gcm_key
    *	x1: gcm_table
    */
+  CFI_STARTPROC()
 
   GET_DATA_POINTER(x2, .Lrconst)
 
@@ -408,6 +420,7 @@ _gcry_ghash_setup_armv8_ce_pmull:
   st1 {rh5.16b-rh6.16b}, [x1]
 
   ret
+  CFI_ENDPROC()
 ELF(.size _gcry_ghash_setup_armv8_ce_pmull,.-_gcry_ghash_setup_armv8_ce_pmull;)
 
 #endif
diff --git a/cipher/crc-armv8-aarch64-ce.S b/cipher/crc-armv8-aarch64-ce.S
index 497d0055..f269b74a 100644
--- a/cipher/crc-armv8-aarch64-ce.S
+++ b/cipher/crc-armv8-aarch64-ce.S
@@ -72,6 +72,7 @@ _gcry_crc32r_armv8_ce_bulk:
    *    x2: inlen
    *    x3: consts
    */
+  CFI_STARTPROC()
 
   GET_DATA_POINTER(x7, .Lcrc32_constants)
   add x9, x3, #consts_k(5 - 1)
@@ -230,6 +231,7 @@ _gcry_crc32r_armv8_ce_bulk:
   st1 {v0.s}[2], [x0]
 
   ret
+  CFI_ENDPROC()
 ELF(.size _gcry_crc32r_armv8_ce_bulk,.-_gcry_crc32r_armv8_ce_bulk;)
 
 /*
@@ -245,6 +247,7 @@ _gcry_crc32r_armv8_ce_reduction_4:
    *    w1: crc
    *    x2: crc32 constants
    */
+  CFI_STARTPROC()
 
   eor v0.16b, v0.16b, v0.16b
   add x2, x2, #consts_my_p(0)
@@ -261,6 +264,7 @@ _gcry_crc32r_armv8_ce_reduction_4:
   mov w0, v0.s[1]
 
   ret
+  CFI_ENDPROC()
 ELF(.size _gcry_crc32r_armv8_ce_reduction_4,.-_gcry_crc32r_armv8_ce_reduction_4;)
 
 /*
@@ -277,6 +281,7 @@ _gcry_crc32_armv8_ce_bulk:
    *    x2: inlen
    *    x3: consts
    */
+  CFI_STARTPROC()
 
   GET_DATA_POINTER(x7, .Lcrc32_constants)
   add x4, x7, #.Lcrc32_bswap_shuf - .Lcrc32_constants
@@ -456,6 +461,7 @@ _gcry_crc32_armv8_ce_bulk:
   st1 {v0.s}[0], [x0]
 
   ret
+  CFI_ENDPROC()
 ELF(.size _gcry_crc32_armv8_ce_bulk,.-_gcry_crc32_armv8_ce_bulk;)
 
 /*
@@ -471,6 +477,7 @@ _gcry_crc32_armv8_ce_reduction_4:
    *    w1: crc
    *    x2: crc32 constants
    */
+  CFI_STARTPROC()
 
   eor v0.16b, v0.16b, v0.16b
   add x2, x2, #consts_my_p(0)
@@ -487,6 +494,7 @@ _gcry_crc32_armv8_ce_reduction_4:
   mov w0, v0.s[0]
 
   ret
+  CFI_ENDPROC()
 ELF(.size _gcry_crc32_armv8_ce_reduction_4,.-_gcry_crc32_armv8_ce_reduction_4;)
 
 #endif
diff --git a/cipher/rijndael-aarch64.S b/cipher/rijndael-aarch64.S
index aad74875..e77dd4e0 100644
--- a/cipher/rijndael-aarch64.S
+++ b/cipher/rijndael-aarch64.S
@@ -216,6 +216,7 @@ _gcry_aes_arm_encrypt_block:
 	 *	%w3: number of rounds.. 10, 12 or 14
 	 *      %x4: encryption table
 	 */
+	CFI_STARTPROC();
 
 	/* read input block */
 
@@ -285,6 +286,7 @@ _gcry_aes_arm_encrypt_block:
 	lastencround(11, RNA, RNB, RNC, RND, RA, RB, RC, RD);
 
 	b .Lenc_done;
+	CFI_ENDPROC();
 ELF(.size _gcry_aes_arm_encrypt_block,.-_gcry_aes_arm_encrypt_block;)
 
 #define addroundkey_dec(round, ra, rb, rc, rd, rna, rnb, rnc, rnd) \
@@ -439,6 +441,7 @@ _gcry_aes_arm_decrypt_block:
 	 *	%w3: number of rounds.. 10, 12 or 14
 	 *      %x4: decryption table
 	 */
+	CFI_STARTPROC();
 
 	/* read input block */
 
@@ -504,6 +507,7 @@ _gcry_aes_arm_decrypt_block:
 	decround(9, RA, RB, RC, RD, RNA, RNB, RNC, RND, preload_first_key);
 
 	b .Ldec_tail;
+	CFI_ENDPROC();
 ELF(.size _gcry_aes_arm_decrypt_block,.-_gcry_aes_arm_decrypt_block;)
 
 #endif /*HAVE_COMPATIBLE_GCC_AARCH64_PLATFORM_AS*/
diff --git a/cipher/rijndael-armv8-aarch64-ce.S b/cipher/rijndael-armv8-aarch64-ce.S
index f0012c20..71b45b85 100644
--- a/cipher/rijndael-armv8-aarch64-ce.S
+++ b/cipher/rijndael-armv8-aarch64-ce.S
@@ -247,6 +247,7 @@ _gcry_aes_enc_armv8_ce:
    *    x2: src
    *    w3: nrounds
    */
+  CFI_STARTPROC();
 
   aes_preload_keys(x0, w3);
 
@@ -291,6 +292,7 @@ _gcry_aes_enc_armv8_ce:
   CLEAR_REG(vk13)
   CLEAR_REG(vk14)
   b .Lenc1_tail
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_enc_armv8_ce,.-_gcry_aes_enc_armv8_ce;)
 
 
@@ -309,6 +311,7 @@ _gcry_aes_dec_armv8_ce:
    *    x2: src
    *    w3: nrounds
    */
+  CFI_STARTPROC();
 
   aes_preload_keys(x0, w3);
 
@@ -353,6 +356,7 @@ _gcry_aes_dec_armv8_ce:
   CLEAR_REG(vk13)
   CLEAR_REG(vk14)
   b .Ldec1_tail
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_dec_armv8_ce,.-_gcry_aes_dec_armv8_ce;)
 
 
@@ -377,6 +381,7 @@ _gcry_aes_cbc_enc_armv8_ce:
    *    w5: cbc_mac
    *    w6: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lcbc_enc_skip
 
@@ -419,6 +424,7 @@ _gcry_aes_cbc_enc_armv8_ce:
 
 .Lcbc_enc_skip:
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_cbc_enc_armv8_ce,.-_gcry_aes_cbc_enc_armv8_ce;)
 
 /*
@@ -440,6 +446,7 @@ _gcry_aes_cbc_dec_armv8_ce:
    *    x4: nblocks
    *    w5: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lcbc_dec_skip
 
@@ -515,6 +522,7 @@ _gcry_aes_cbc_dec_armv8_ce:
 
 .Lcbc_dec_skip:
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_cbc_dec_armv8_ce,.-_gcry_aes_cbc_dec_armv8_ce;)
 
 
@@ -537,6 +545,7 @@ _gcry_aes_ctr_enc_armv8_ce:
    *    x4: nblocks
    *    w5: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lctr_enc_skip
 
@@ -668,7 +677,7 @@ _gcry_aes_ctr_enc_armv8_ce:
 
 .Lctr_enc_skip:
   ret
-
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_ctr_enc_armv8_ce,.-_gcry_aes_ctr_enc_armv8_ce;)
 
 
@@ -691,6 +700,7 @@ _gcry_aes_cfb_enc_armv8_ce:
    *    x4: nblocks
    *    w5: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lcfb_enc_skip
 
@@ -732,6 +742,7 @@ _gcry_aes_cfb_enc_armv8_ce:
 
 .Lcfb_enc_skip:
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_cfb_enc_armv8_ce,.-_gcry_aes_cfb_enc_armv8_ce;)
 
 
@@ -754,6 +765,7 @@ _gcry_aes_cfb_dec_armv8_ce:
    *    x4: nblocks
    *    w5: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lcfb_dec_skip
 
@@ -829,6 +841,7 @@ _gcry_aes_cfb_dec_armv8_ce:
 
 .Lcfb_dec_skip:
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_cfb_dec_armv8_ce,.-_gcry_aes_cfb_dec_armv8_ce;)
 
 
@@ -859,6 +872,7 @@ _gcry_aes_ocb_enc_armv8_ce:
    *    w7: nrounds
    *    %st+0: blkn => w12
    */
+  CFI_STARTPROC();
 
   ldr w12, [sp]
   ld1 {v0.16b}, [x3] /* load offset */
@@ -979,6 +993,7 @@ _gcry_aes_ocb_enc_armv8_ce:
   CLEAR_REG(v16)
 
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_ocb_enc_armv8_ce,.-_gcry_aes_ocb_enc_armv8_ce;)
 
 
@@ -1009,6 +1024,7 @@ _gcry_aes_ocb_dec_armv8_ce:
    *    w7: nrounds
    *    %st+0: blkn => w12
    */
+  CFI_STARTPROC();
 
   ldr w12, [sp]
   ld1 {v0.16b}, [x3] /* load offset */
@@ -1129,6 +1145,7 @@ _gcry_aes_ocb_dec_armv8_ce:
   CLEAR_REG(v16)
 
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_ocb_dec_armv8_ce,.-_gcry_aes_ocb_dec_armv8_ce;)
 
 
@@ -1157,6 +1174,8 @@ _gcry_aes_ocb_auth_armv8_ce:
    *    w6: nrounds => w7
    *    w7: blkn => w12
    */
+  CFI_STARTPROC();
+
   mov w12, w7
   mov w7, w6
   mov x6, x5
@@ -1273,6 +1292,7 @@ _gcry_aes_ocb_auth_armv8_ce:
   CLEAR_REG(v16)
 
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_ocb_auth_armv8_ce,.-_gcry_aes_ocb_auth_armv8_ce;)
 
 
@@ -1297,6 +1317,7 @@ _gcry_aes_xts_enc_armv8_ce:
    *    x4: nblocks
    *    w5: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lxts_enc_skip
 
@@ -1411,7 +1432,7 @@ _gcry_aes_xts_enc_armv8_ce:
 
 .Lxts_enc_skip:
   ret
-
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_xts_enc_armv8_ce,.-_gcry_aes_xts_enc_armv8_ce;)
 
 
@@ -1436,6 +1457,7 @@ _gcry_aes_xts_dec_armv8_ce:
    *    x4: nblocks
    *    w5: nrounds
    */
+  CFI_STARTPROC();
 
   cbz x4, .Lxts_dec_skip
 
@@ -1550,7 +1572,7 @@ _gcry_aes_xts_dec_armv8_ce:
 
 .Lxts_dec_skip:
   ret
-
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_xts_dec_armv8_ce,.-_gcry_aes_xts_dec_armv8_ce;)
 
 
@@ -1564,6 +1586,7 @@ _gcry_aes_sbox4_armv8_ce:
   /* See "Gouvêa, C. P. L. & López, J. Implementing GCM on ARMv8. Topics in
    * Cryptology — CT-RSA 2015" for details.
    */
+  CFI_STARTPROC();
   movi v0.16b, #0x52
   movi v1.16b, #0
   mov v0.S[0], w0
@@ -1572,6 +1595,7 @@ _gcry_aes_sbox4_armv8_ce:
   mov w0, v0.S[0]
   CLEAR_REG(v0)
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_sbox4_armv8_ce,.-_gcry_aes_sbox4_armv8_ce;)
 
 
@@ -1582,11 +1606,13 @@ ELF(.size _gcry_aes_sbox4_armv8_ce,.-_gcry_aes_sbox4_armv8_ce;)
 .globl _gcry_aes_invmixcol_armv8_ce
 ELF(.type  _gcry_aes_invmixcol_armv8_ce,%function;)
 _gcry_aes_invmixcol_armv8_ce:
+  CFI_STARTPROC();
   ld1 {v0.16b}, [x1]
   aesimc v0.16b, v0.16b
   st1 {v0.16b}, [x0]
   CLEAR_REG(v0)
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_aes_invmixcol_armv8_ce,.-_gcry_aes_invmixcol_armv8_ce;)
 
 #endif
diff --git a/cipher/sha1-armv8-aarch64-ce.S b/cipher/sha1-armv8-aarch64-ce.S
index aeb67a12..7dc26c0f 100644
--- a/cipher/sha1-armv8-aarch64-ce.S
+++ b/cipher/sha1-armv8-aarch64-ce.S
@@ -110,6 +110,7 @@ _gcry_sha1_transform_armv8_ce:
    *	x1: data (64*nblks bytes)
    *	x2: nblks
    */
+  CFI_STARTPROC();
 
   cbz x2, .Ldo_nothing;
 
@@ -199,6 +200,7 @@ _gcry_sha1_transform_armv8_ce:
 .Ldo_nothing:
   mov x0, #0
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_sha1_transform_armv8_ce,.-_gcry_sha1_transform_armv8_ce;)
 
 #endif
diff --git a/cipher/sha256-armv8-aarch64-ce.S b/cipher/sha256-armv8-aarch64-ce.S
index 6b3ad32d..706e0dfd 100644
--- a/cipher/sha256-armv8-aarch64-ce.S
+++ b/cipher/sha256-armv8-aarch64-ce.S
@@ -120,6 +120,7 @@ _gcry_sha256_transform_armv8_ce:
    *	r1: data (64*nblks bytes)
    *	r2: nblks
    */
+  CFI_STARTPROC();
 
   cbz x2, .Ldo_nothing;
 
@@ -213,6 +214,7 @@ _gcry_sha256_transform_armv8_ce:
 .Ldo_nothing:
   mov x0, #0
   ret
+  CFI_ENDPROC();
 ELF(.size _gcry_sha256_transform_armv8_ce,.-_gcry_sha256_transform_armv8_ce;)
 
 #endif
diff --git a/cipher/twofish-aarch64.S b/cipher/twofish-aarch64.S
index adee412d..9f35b5cd 100644
--- a/cipher/twofish-aarch64.S
+++ b/cipher/twofish-aarch64.S
@@ -225,6 +225,7 @@ _gcry_twofish_arm_encrypt_block:
 	 *	x1: dst
 	 *	x2: src
 	 */
+	CFI_STARTPROC();
 
 	add CTXw, CTX, #(w);
 
@@ -262,6 +263,7 @@ _gcry_twofish_arm_encrypt_block:
 	str_output_le(RDST, RC, RD, RA, RB, RT0, RT1);
 
 	ret;
+	CFI_ENDPROC();
 .ltorg
 ELF(.size _gcry_twofish_arm_encrypt_block,.-_gcry_twofish_arm_encrypt_block;)
 
@@ -274,6 +276,7 @@ _gcry_twofish_arm_decrypt_block:
 	 *	%r1: dst
 	 *	%r2: src
 	 */
+	CFI_STARTPROC();
 
 	add CTXw, CTX, #(w);
 
@@ -311,6 +314,7 @@ _gcry_twofish_arm_decrypt_block:
 	str_output_le(RDST, RA, RB, RC, RD, RT0, RT1);
 
 	ret;
+	CFI_ENDPROC();
 ELF(.size _gcry_twofish_arm_decrypt_block,.-_gcry_twofish_arm_decrypt_block;)
 
 #endif /*HAVE_COMPATIBLE_GCC_AARCH64_PLATFORM_AS*/
diff --git a/mpi/aarch64/mpih-add1.S b/mpi/aarch64/mpih-add1.S
index 3370320e..bc62cf98 100644
--- a/mpi/aarch64/mpih-add1.S
+++ b/mpi/aarch64/mpih-add1.S
@@ -37,6 +37,7 @@
 .globl _gcry_mpih_add_n
 ELF(.type  _gcry_mpih_add_n,%function)
 _gcry_mpih_add_n:
+	CFI_STARTPROC()
 	and	w5, w3, #3;
 	adds	xzr, xzr, xzr; /* clear carry flag */
 
@@ -69,4 +70,5 @@ _gcry_mpih_add_n:
 .Lend:
 	adc	x0, xzr, xzr;
 	ret;
+	CFI_ENDPROC()
 ELF(.size _gcry_mpih_add_n,.-_gcry_mpih_add_n;)
diff --git a/mpi/aarch64/mpih-mul1.S b/mpi/aarch64/mpih-mul1.S
index 8830845a..92fcd141 100644
--- a/mpi/aarch64/mpih-mul1.S
+++ b/mpi/aarch64/mpih-mul1.S
@@ -37,6 +37,7 @@
 .globl _gcry_mpih_mul_1
 ELF(.type  _gcry_mpih_mul_1,%function)
 _gcry_mpih_mul_1:
+	CFI_STARTPROC()
 	and	w5, w2, #3;
 	mov	x4, xzr;
 
@@ -94,4 +95,5 @@ _gcry_mpih_mul_1:
 .Lend:
 	mov	x0, x4;
 	ret;
+	CFI_ENDPROC()
 ELF(.size _gcry_mpih_mul_1,.-_gcry_mpih_mul_1;)
diff --git a/mpi/aarch64/mpih-mul2.S b/mpi/aarch64/mpih-mul2.S
index 5d736990..aa0e5a2d 100644
--- a/mpi/aarch64/mpih-mul2.S
+++ b/mpi/aarch64/mpih-mul2.S
@@ -37,6 +37,7 @@
 .globl _gcry_mpih_addmul_1
 ELF(.type  _gcry_mpih_addmul_1,%function)
 _gcry_mpih_addmul_1:
+	CFI_STARTPROC()
 	and	w5, w2, #3;
 	mov	x6, xzr;
 	mov	x7, xzr;
@@ -106,4 +107,5 @@ _gcry_mpih_addmul_1:
 .Lend:
 	mov	x0, x6;
 	ret;
+	CFI_ENDPROC()
 ELF(.size _gcry_mpih_addmul_1,.-_gcry_mpih_addmul_1;)
diff --git a/mpi/aarch64/mpih-mul3.S b/mpi/aarch64/mpih-mul3.S
index f785e5e4..5a40b354 100644
--- a/mpi/aarch64/mpih-mul3.S
+++ b/mpi/aarch64/mpih-mul3.S
@@ -37,6 +37,7 @@
 .globl _gcry_mpih_submul_1
 ELF(.type  _gcry_mpih_submul_1,%function)
 _gcry_mpih_submul_1:
+	CFI_STARTPROC()
 	and	w5, w2, #3;
 	mov	x7, xzr;
 	cbz	w5, .Large_loop;
@@ -119,4 +120,5 @@ _gcry_mpih_submul_1:
 .Loop_end:
 	cinc	x0, x7, cc;
 	ret;
+	CFI_ENDPROC()
 ELF(.size _gcry_mpih_submul_1,.-_gcry_mpih_submul_1;)
diff --git a/mpi/aarch64/mpih-sub1.S b/mpi/aarch64/mpih-sub1.S
index 45a7b041..4f279a12 100644
--- a/mpi/aarch64/mpih-sub1.S
+++ b/mpi/aarch64/mpih-sub1.S
@@ -37,6 +37,7 @@
 .globl _gcry_mpih_sub_n
 ELF(.type  _gcry_mpih_sub_n,%function)
 _gcry_mpih_sub_n:
+	CFI_STARTPROC()
 	and	w5, w3, #3;
 	subs	xzr, xzr, xzr; /* prepare carry flag for sub */
 
@@ -69,4 +70,5 @@ _gcry_mpih_sub_n:
 .Lend:
 	cset	x0, cc;
 	ret;
+	CFI_ENDPROC()
 ELF(.size _gcry_mpih_sub_n,.-_gcry_mpih_sub_n;)
diff --git a/mpi/asm-common-aarch64.h b/mpi/asm-common-aarch64.h
index 12694130..cf4bdb85 100644
--- a/mpi/asm-common-aarch64.h
+++ b/mpi/asm-common-aarch64.h
@@ -21,10 +21,6 @@
 #ifndef MPI_ASM_COMMON_AARCH64_H
 #define MPI_ASM_COMMON_AARCH64_H
 
-#ifdef __ELF__
-# define ELF(...) __VA_ARGS__
-#else
-# define ELF(...) /*_*/
-#endif
+#include "../cipher/asm-common-aarch64.h"
 
 #endif /* MPI_ASM_COMMON_AARCH64_H */