AArch64: Fix assertions in processing dynamic relocations

This patch provides the following changes in fixup_gdt_reloc()
function:
- Fixes assertions in processing dynamic relocations, when
relocation entries not matching R_AARCH64_RELATIVE type are found.
Linker might generate entries of relocation type R_AARCH64_NONE
(code 0), which should be ignored to make the code boot. Similar
issue was fixed in OP-TEE (see optee_os/ldelf/ta_elf_rel.c
commit 7a4dc765c133125428136a496a7644c6fec9b3c2)
- Fixes bug when "b.ge" (signed greater than or equal) condition
codes were used instead of "b.hs" (greater than or equal) for
comparison of absolute addresses.
- Adds optimisation which skips fixing Global Object Table (GOT)
entries when offset value is 0.

Change-Id: I35e34e055b7476843903859be947b883a1feb1b5
Signed-off-by: Alexei Fedorov <Alexei.Fedorov@arm.com>

lib/aarch64/misc_helpers.S

 /*
- * Copyright (c) 2013-2020, ARM Limited and Contributors. All rights reserved.
+ * Copyright (c) 2013-2021, Arm Limited and Contributors. All rights reserved.
  *
  * SPDX-License-Identifier: BSD-3-Clause
  */
@@ -486,15 +486,20 @@ endfunc enable_vfp
  * arguments (which is usually the limits of the relocable BL image).
  *   x0 -  the start of the fixup region
  *   x1 -  the limit of the fixup region
- * These addresses have to be page (4KB aligned).
+ * These addresses have to be 4KB page aligned.
  * ---------------------------------------------------------------------------
  */
+
+/* Relocation codes */
+#define	R_AARCH64_NONE		0
+#define	R_AARCH64_RELATIVE	1027
+
 func fixup_gdt_reloc
 	mov	x6, x0
 	mov	x7, x1
 
-	/* Test if the limits are 4K aligned */
 #if ENABLE_ASSERTIONS
+	/* Test if the limits are 4KB aligned */
 	orr	x0, x0, x1
 	tst	x0, #(PAGE_SIZE_MASK)
 	ASM_ASSERT(eq)
@@ -505,7 +510,8 @@ func fixup_gdt_reloc
 	 * fixup region.
 	 */
 	and	x2, x30, #~(PAGE_SIZE_MASK)
-	sub	x0, x2, x6	/* Diff(S) = Current Address - Compiled Address */
+	subs	x0, x2, x6	/* Diff(S) = Current Address - Compiled Address */
+	b.eq	3f		/* Diff(S) = 0. No relocation needed */
 
 	adrp	x1, __GOT_START__
 	add	x1, x1, :lo12:__GOT_START__
@@ -518,31 +524,32 @@ func fixup_gdt_reloc
 	 * The new_addr is the address currently the binary is executing from
 	 * and old_addr is the address at compile time.
 	 */
-1:
-	ldr	x3, [x1]
+1:	ldr	x3, [x1]
+
 	/* Skip adding offset if address is < lower limit */
 	cmp	x3, x6
 	b.lo	2f
+
 	/* Skip adding offset if address is >= upper limit */
 	cmp	x3, x7
-	b.ge	2f
+	b.hs	2f
 	add	x3, x3, x0
 	str	x3, [x1]
-2:
-	add	x1, x1, #8
+
+2:	add	x1, x1, #8
 	cmp	x1, x2
 	b.lo	1b
 
 	/* Starting dynamic relocations. Use adrp/adr to get RELA_START and END */
-	adrp	x1, __RELA_START__
+3:	adrp	x1, __RELA_START__
 	add	x1, x1, :lo12:__RELA_START__
 	adrp	x2, __RELA_END__
 	add	x2, x2, :lo12:__RELA_END__
+
 	/*
 	 * According to ELF-64 specification, the RELA data structure is as
 	 * follows:
-	 *	typedef struct
-	 * 	{
+	 *	typedef struct {
 	 *		Elf64_Addr r_offset;
 	 *		Elf64_Xword r_info;
 	 *		Elf64_Sxword r_addend;
@@ -550,16 +557,19 @@ func fixup_gdt_reloc
 	 *
 	 * r_offset is address of reference
 	 * r_info is symbol index and type of relocation (in this case
-	 * 0x403 which corresponds to R_AARCH64_RELATIVE).
+	 * code 1027 which corresponds to R_AARCH64_RELATIVE).
 	 * r_addend is constant part of expression.
 	 *
 	 * Size of Elf64_Rela structure is 24 bytes.
 	 */
-1:
-	/* Assert that the relocation type is R_AARCH64_RELATIVE */
+
+	/* Skip R_AARCH64_NONE entry with code 0 */
+1:	ldr	x3, [x1, #8]
+	cbz	x3, 2f
+
 #if ENABLE_ASSERTIONS
-	ldr	x3, [x1, #8]
-	cmp	x3, #0x403
+	/* Assert that the relocation type is R_AARCH64_RELATIVE */
+	cmp	x3, #R_AARCH64_RELATIVE
 	ASM_ASSERT(eq)
 #endif
 	ldr	x3, [x1]	/* r_offset */
@@ -569,9 +579,10 @@ func fixup_gdt_reloc
 	/* Skip adding offset if r_addend is < lower limit */
 	cmp	x4, x6
 	b.lo	2f
+
 	/* Skip adding offset if r_addend entry is >= upper limit */
 	cmp	x4, x7
-	b.ge	2f
+	b.hs	2f
 
 	add	x4, x0, x4	/* Diff(S) + r_addend */
 	str	x4, [x3]
@@ -579,6 +590,5 @@ func fixup_gdt_reloc
 2:	add	x1, x1, #24
 	cmp	x1, x2
 	b.lo	1b
-
 	ret
 endfunc fixup_gdt_reloc