aarch64-asm: use ADR for getting pointers for local labels

* cipher/asm-common-aarch64.h (GET_DATA_POINTER): Remove.
(GET_LOCAL_POINTER): New.
* cipher/camellia-aarch64.S: Use GET_LOCAL_POINTER instead of ADR
instruction directly.
* cipher/chacha20-aarch64.S: Use GET_LOCAL_POINTER instead of
GET_DATA_POINTER.
* cipher/cipher-gcm-armv8-aarch64-ce.S: Likewise.
* cipher/crc-armv8-aarch64-ce.S: Likewise.
* cipher/sha1-armv8-aarch64-ce.S: Likewise.
* cipher/sha256-armv8-aarch64-ce.S: Likewise.
* cipher/sm3-aarch64.S: Likewise.
* cipher/sm3-armv8-aarch64-ce.S: Likewise.
* cipher/sm4-aarch64.S: Likewise.
---

Switch to use ADR instead of ADRP/LDR or ADRP/ADD for getting
data pointers within assembly files. ADR is more portable across
targets and does not require labels to be declared in GOT tables.

Reviewed-and-tested-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

10 files changed, 18 insertions, 29 deletions

diff --git a/cipher/asm-common-aarch64.h b/cipher/asm-common-aarch64.h
index d3f7801c..b38b17a6 100644
--- a/cipher/asm-common-aarch64.h
+++ b/cipher/asm-common-aarch64.h
@@ -29,19 +29,8 @@
 # define ELF(...) /*_*/
 #endif
 
-#ifdef __APPLE__
-#define GET_DATA_POINTER(reg, name) \
-	adrp    reg, name@GOTPAGE ; \
-	add     reg, reg, name@GOTPAGEOFF ;
-#elif defined(_WIN32)
-#define GET_DATA_POINTER(reg, name) \
-	adrp    reg, name ; \
-	add     reg, reg, #:lo12:name ;
-#else
-#define GET_DATA_POINTER(reg, name) \
-	adrp    reg, :got:name ; \
-	ldr     reg, [reg, #:got_lo12:name] ;
-#endif
+#define GET_LOCAL_POINTER(reg, label) \
+	adr reg, label;
 
 #ifdef HAVE_GCC_ASM_CFI_DIRECTIVES
 /* CFI directives to emit DWARF stack unwinding information. */
diff --git a/cipher/camellia-aarch64.S b/cipher/camellia-aarch64.S
index 30b568d3..c019c168 100644
--- a/cipher/camellia-aarch64.S
+++ b/cipher/camellia-aarch64.S
@@ -214,7 +214,7 @@ _gcry_camellia_arm_encrypt_block:
 	 *	w3: keybitlen
 	 */
 
-	adr RTAB1,  _gcry_camellia_arm_tables;
+	GET_LOCAL_POINTER(RTAB1,  _gcry_camellia_arm_tables);
 	mov RMASK, #(0xff<<4); /* byte mask */
 	add RTAB2, RTAB1, #(1 * 4);
 	add RTAB3, RTAB1, #(2 * 4);
@@ -274,7 +274,7 @@ _gcry_camellia_arm_decrypt_block:
 	 *	w3: keybitlen
 	 */
 
-	adr RTAB1,  _gcry_camellia_arm_tables;
+	GET_LOCAL_POINTER(RTAB1,  _gcry_camellia_arm_tables);
 	mov RMASK, #(0xff<<4); /* byte mask */
 	add RTAB2, RTAB1, #(1 * 4);
 	add RTAB3, RTAB1, #(2 * 4);
diff --git a/cipher/chacha20-aarch64.S b/cipher/chacha20-aarch64.S
index 2a980b95..540f892b 100644
--- a/cipher/chacha20-aarch64.S
+++ b/cipher/chacha20-aarch64.S
@@ -206,10 +206,10 @@ _gcry_chacha20_aarch64_blocks4:
 	 */
 	CFI_STARTPROC()
 
-	GET_DATA_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_rot8);
+	GET_LOCAL_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_rot8);
 	add INPUT_CTR, INPUT, #(12*4);
 	ld1 {ROT8.16b}, [CTR];
-	GET_DATA_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_inc_counter);
+	GET_LOCAL_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_inc_counter);
 	mov INPUT_POS, INPUT;
 	ld1 {VCTR.16b}, [CTR];
 
@@ -383,10 +383,10 @@ _gcry_chacha20_poly1305_aarch64_blocks4:
 	mov POLY_RSTATE, x4;
 	mov POLY_RSRC, x5;
 
-	GET_DATA_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_rot8);
+	GET_LOCAL_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_rot8);
 	add INPUT_CTR, INPUT, #(12*4);
 	ld1 {ROT8.16b}, [CTR];
-	GET_DATA_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_inc_counter);
+	GET_LOCAL_POINTER(CTR, _gcry_chacha20_aarch64_blocks4_data_inc_counter);
 	mov INPUT_POS, INPUT;
 	ld1 {VCTR.16b}, [CTR];
 
diff --git a/cipher/cipher-gcm-armv8-aarch64-ce.S b/cipher/cipher-gcm-armv8-aarch64-ce.S
index 687fabe3..78f3ad2d 100644
--- a/cipher/cipher-gcm-armv8-aarch64-ce.S
+++ b/cipher/cipher-gcm-armv8-aarch64-ce.S
@@ -169,7 +169,7 @@ _gcry_ghash_armv8_ce_pmull:
 
   cbz x3, .Ldo_nothing;
 
-  GET_DATA_POINTER(x5, .Lrconst)
+  GET_LOCAL_POINTER(x5, .Lrconst)
 
   eor vZZ.16b, vZZ.16b, vZZ.16b
   ld1 {rhash.16b}, [x1]
@@ -368,7 +368,7 @@ _gcry_polyval_armv8_ce_pmull:
 
   cbz x3, .Lpolyval_do_nothing;
 
-  GET_DATA_POINTER(x5, .Lrconst)
+  GET_LOCAL_POINTER(x5, .Lrconst)
 
   eor vZZ.16b, vZZ.16b, vZZ.16b
   ld1 {rhash.16b}, [x1]
@@ -589,7 +589,7 @@ _gcry_ghash_setup_armv8_ce_pmull:
    */
   CFI_STARTPROC()
 
-  GET_DATA_POINTER(x2, .Lrconst)
+  GET_LOCAL_POINTER(x2, .Lrconst)
 
   eor vZZ.16b, vZZ.16b, vZZ.16b
 
diff --git a/cipher/crc-armv8-aarch64-ce.S b/cipher/crc-armv8-aarch64-ce.S
index 7ac884af..b6cdbb3d 100644
--- a/cipher/crc-armv8-aarch64-ce.S
+++ b/cipher/crc-armv8-aarch64-ce.S
@@ -71,7 +71,7 @@ _gcry_crc32r_armv8_ce_bulk:
    */
   CFI_STARTPROC()
 
-  GET_DATA_POINTER(x7, .Lcrc32_constants)
+  GET_LOCAL_POINTER(x7, .Lcrc32_constants)
   add x9, x3, #consts_k(5 - 1)
   cmp x2, #128
 
@@ -280,7 +280,7 @@ _gcry_crc32_armv8_ce_bulk:
    */
   CFI_STARTPROC()
 
-  GET_DATA_POINTER(x7, .Lcrc32_constants)
+  GET_LOCAL_POINTER(x7, .Lcrc32_constants)
   add x4, x7, #.Lcrc32_bswap_shuf - .Lcrc32_constants
   cmp x2, #128
   ld1 {v7.16b}, [x4]
diff --git a/cipher/sha1-armv8-aarch64-ce.S b/cipher/sha1-armv8-aarch64-ce.S
index ea26564b..f95717ee 100644
--- a/cipher/sha1-armv8-aarch64-ce.S
+++ b/cipher/sha1-armv8-aarch64-ce.S
@@ -109,7 +109,7 @@ _gcry_sha1_transform_armv8_ce:
 
   cbz x2, .Ldo_nothing;
 
-  GET_DATA_POINTER(x4, .LK_VEC);
+  GET_LOCAL_POINTER(x4, .LK_VEC);
 
   ld1 {vH0123.4s}, [x0]     /* load h0,h1,h2,h3 */
   ld1 {vK1.4s-vK4.4s}, [x4] /* load K1,K2,K3,K4 */
diff --git a/cipher/sha256-armv8-aarch64-ce.S b/cipher/sha256-armv8-aarch64-ce.S
index d0fa6285..5616eada 100644
--- a/cipher/sha256-armv8-aarch64-ce.S
+++ b/cipher/sha256-armv8-aarch64-ce.S
@@ -119,7 +119,7 @@ _gcry_sha256_transform_armv8_ce:
 
   cbz x2, .Ldo_nothing;
 
-  GET_DATA_POINTER(x3, .LK);
+  GET_LOCAL_POINTER(x3, .LK);
   mov x4, x3
 
   ld1 {vH0123.4s-vH4567.4s}, [x0]  /* load state */
diff --git a/cipher/sm3-aarch64.S b/cipher/sm3-aarch64.S
index 3fb89006..0e58254b 100644
--- a/cipher/sm3-aarch64.S
+++ b/cipher/sm3-aarch64.S
@@ -425,7 +425,7 @@ _gcry_sm3_transform_aarch64:
   CFI_DEF_CFA_REGISTER(RFRAME);
 
   sub addr0, sp, #STACK_SIZE;
-  GET_DATA_POINTER(RKPTR, .LKtable);
+  GET_LOCAL_POINTER(RKPTR, .LKtable);
   and sp, addr0, #(~63);
 
   /* Preload first block. */
diff --git a/cipher/sm3-armv8-aarch64-ce.S b/cipher/sm3-armv8-aarch64-ce.S
index 0900b84f..d592d08a 100644
--- a/cipher/sm3-armv8-aarch64-ce.S
+++ b/cipher/sm3-armv8-aarch64-ce.S
@@ -170,7 +170,7 @@ _gcry_sm3_transform_armv8_ce:
     ext         CTX2.16b, CTX2.16b, CTX2.16b, #8;
 
 .Lloop:
-    GET_DATA_POINTER(x3, .Lsm3_Ktable);
+    GET_LOCAL_POINTER(x3, .Lsm3_Ktable);
     ld1         {v0.16b-v3.16b}, [x1], #64;
     sub         x2, x2, #1;
 
diff --git a/cipher/sm4-aarch64.S b/cipher/sm4-aarch64.S
index 306b425e..8d06991b 100644
--- a/cipher/sm4-aarch64.S
+++ b/cipher/sm4-aarch64.S
@@ -84,7 +84,7 @@ ELF(.size _gcry_sm4_aarch64_consts,.-_gcry_sm4_aarch64_consts)
 /* Helper macros. */
 
 #define preload_sbox(ptr)                   \
-        GET_DATA_POINTER(ptr, .Lsm4_sbox);  \
+        GET_LOCAL_POINTER(ptr, .Lsm4_sbox);  \
         ld1 {v16.16b-v19.16b}, [ptr], #64;  \
         ld1 {v20.16b-v23.16b}, [ptr], #64;  \
         ld1 {v24.16b-v27.16b}, [ptr], #64;  \