[PATCH v4 09/15] ARM: module: implement support for PC-relative group relocations

[Ard Biesheuvel <ardb@kernel.org>]

Add support for the R_ARM_ALU_PC_Gn_NC and R_ARM_LDR_PC_G2 group
relocations [0] so we can use them in modules. These will be used to
load the current task pointer from a global variable without having to
rely on a literal pool entry to carry the address of this variable,
which may have a significant negative impact on cache utilization for
variables that are used often and in many different places, as each
occurrence will result in a literal pool entry and therefore a line in
the D-cache.

[0] 'ELF for the ARM architecture'
    https://github.com/ARM-software/abi-aa/releases

Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Marc Zyngier <maz@kernel.org>
Tested-by: Vladimir Murzin <vladimir.murzin@arm.com> # ARMv7M
---
 arch/arm/include/asm/elf.h |  3 +
 arch/arm/kernel/module.c   | 85 ++++++++++++++++++++
 2 files changed, 88 insertions(+)

diff --git a/arch/arm/include/asm/elf.h b/arch/arm/include/asm/elf.h
index b8102a6ddf16..d68101655b74 100644
--- a/arch/arm/include/asm/elf.h
+++ b/arch/arm/include/asm/elf.h
@@ -61,6 +61,9 @@ typedef struct user_fp elf_fpregset_t;
 #define R_ARM_MOVT_ABS		44
 #define R_ARM_MOVW_PREL_NC	45
 #define R_ARM_MOVT_PREL		46
+#define R_ARM_ALU_PC_G0_NC	57
+#define R_ARM_ALU_PC_G1_NC	59
+#define R_ARM_LDR_PC_G2		63
 
 #define R_ARM_THM_CALL		10
 #define R_ARM_THM_JUMP24	30
diff --git a/arch/arm/kernel/module.c b/arch/arm/kernel/module.c
index beac45e89ba6..4d33a7acf617 100644
--- a/arch/arm/kernel/module.c
+++ b/arch/arm/kernel/module.c
@@ -68,6 +68,42 @@ bool module_exit_section(const char *name)
 		strstarts(name, ".ARM.exidx.exit");
 }
 
+/*
+ * This implements the partitioning algorithm for group relocations as
+ * documented in the ARM AArch32 ELF psABI (IHI 0044).
+ *
+ * A single PC-relative symbol reference is divided in up to 3 add or subtract
+ * operations, where the final one could be incorporated into a load/store
+ * instruction with immediate offset. E.g.,
+ *
+ *   ADD	Rd, PC, #...		or	ADD	Rd, PC, #...
+ *   ADD	Rd, Rd, #...			ADD	Rd, Rd, #...
+ *   LDR	Rd, [Rd, #...]			ADD	Rd, Rd, #...
+ *
+ * The latter has a guaranteed range of only 16 MiB (3x8 == 24 bits), so it is
+ * of limited use in the kernel. However, the ADD/ADD/LDR combo has a range of
+ * -/+ 256 MiB, (2x8 + 12 == 28 bits), which means it has sufficient range for
+ * any in-kernel symbol reference (unless module PLTs are being used).
+ *
+ * The main advantage of this approach over the typical pattern using a literal
+ * load is that literal loads may miss in the D-cache, and generally lead to
+ * lower cache efficiency for variables that are referenced often from many
+ * different places in the code.
+ */
+static u32 get_group_rem(u32 group, u32 *offset)
+{
+	u32 val = *offset;
+	u32 shift;
+	do {
+		shift = val ? (31 - __fls(val)) & ~1 : 32;
+		*offset = val;
+		if (!val)
+			break;
+		val &= 0xffffff >> shift;
+	} while (group--);
+	return shift;
+}
+
 int
 apply_relocate(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex,
 	       unsigned int relindex, struct module *module)
@@ -82,6 +118,7 @@ apply_relocate(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex,
 		unsigned long loc;
 		Elf32_Sym *sym;
 		const char *symname;
+		u32 shift, group = 1;
 		s32 offset;
 		u32 tmp;
 #ifdef CONFIG_THUMB2_KERNEL
@@ -212,6 +249,54 @@ apply_relocate(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex,
 			*(u32 *)loc = __opcode_to_mem_arm(tmp);
 			break;
 
+		case R_ARM_ALU_PC_G0_NC:
+			group = 0;
+			fallthrough;
+		case R_ARM_ALU_PC_G1_NC:
+			tmp = __mem_to_opcode_arm(*(u32 *)loc);
+			offset = ror32(tmp & 0xff, (tmp & 0xf00) >> 7);
+			if (tmp & BIT(22))
+				offset = -offset;
+			offset += sym->st_value - loc;
+			if (offset < 0) {
+				offset = -offset;
+				tmp = (tmp & ~BIT(23)) | BIT(22); // SUB opcode
+			} else {
+				tmp = (tmp & ~BIT(22)) | BIT(23); // ADD opcode
+			}
+
+			shift = get_group_rem(group, &offset);
+			if (shift < 24) {
+				offset >>= 24 - shift;
+				offset |= (shift + 8) << 7;
+			}
+			*(u32 *)loc = __opcode_to_mem_arm((tmp & ~0xfff) | offset);
+			break;
+
+		case R_ARM_LDR_PC_G2:
+			tmp = __mem_to_opcode_arm(*(u32 *)loc);
+			offset = tmp & 0xfff;
+			if (~tmp & BIT(23))		// U bit cleared?
+				offset = -offset;
+			offset += sym->st_value - loc;
+			if (offset < 0) {
+				offset = -offset;
+				tmp &= ~BIT(23);	// clear U bit
+			} else {
+				tmp |= BIT(23);		// set U bit
+			}
+			get_group_rem(2, &offset);
+
+			if (offset > 0xfff) {
+				pr_err("%s: section %u reloc %u sym '%s': relocation %u out of range (%#lx -> %#x)\n",
+				       module->name, relindex, i, symname,
+				       ELF32_R_TYPE(rel->r_info), loc,
+				       sym->st_value);
+				return -ENOEXEC;
+			}
+			*(u32 *)loc = __opcode_to_mem_arm((tmp & ~0xfff) | offset);
+			break;
+
 #ifdef CONFIG_THUMB2_KERNEL
 		case R_ARM_THM_CALL:
 		case R_ARM_THM_JUMP24:
-- 
2.30.2


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel