arm64: add module support for alternatives fixups

Currently the kernel patches all necessary instructions once at boot
time, so modules are not covered by this.
Change the apply_alternatives() function to take a beginning and an
end pointer and introduce a new variant (apply_alternatives_all()) to
cover the existing use case for the static kernel image section.
Add a module_finalize() function to arm64 to check for an
alternatives section in a module and patch only the instructions from
that specific area.
Since that module code is not touched before the module
initialization has ended, we don't need to halt the machine before
doing the patching in the module's code.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

arch/arm64/include/asm/alternative.h

@@ -13,7 +13,8 @@ struct alt_instr {
 	u8  alt_len;		/* size of new instruction(s), <= orig_len */
 };
 
-void apply_alternatives(void);
+void apply_alternatives_all(void);
+void apply_alternatives(void *start, size_t length);
 void free_alternatives_memory(void);
 
 #define ALTINSTR_ENTRY(feature)						      \

arch/arm64/kernel/alternative.c

@@ -28,12 +28,18 @@
 
 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
 
-static int __apply_alternatives(void *dummy)
+struct alt_region {
+	struct alt_instr *begin;
+	struct alt_instr *end;
+};
+
+static int __apply_alternatives(void *alt_region)
 {
 	struct alt_instr *alt;
+	struct alt_region *region = alt_region;
 	u8 *origptr, *replptr;
 
-	for (alt = __alt_instructions; alt < __alt_instructions_end; alt++) {
+	for (alt = region->begin; alt < region->end; alt++) {
 		if (!cpus_have_cap(alt->cpufeature))
 			continue;
 
@@ -51,10 +57,25 @@ static int __apply_alternatives(void *dummy)
 	return 0;
 }
 
-void apply_alternatives(void)
+void apply_alternatives_all(void)
 {
+	struct alt_region region = {
+		.begin	= __alt_instructions,
+		.end	= __alt_instructions_end,
+	};
+
 	/* better not try code patching on a live SMP system */
-	stop_machine(__apply_alternatives, NULL, NULL);
+	stop_machine(__apply_alternatives, &region, NULL);
+}
+
+void apply_alternatives(void *start, size_t length)
+{
+	struct alt_region region = {
+		.begin	= start,
+		.end	= start + length,
+	};
+
+	__apply_alternatives(&region);
 }
 
 void free_alternatives_memory(void)

arch/arm64/kernel/module.c

@@ -26,6 +26,7 @@
 #include <linux/moduleloader.h>
 #include <linux/vmalloc.h>
 #include <asm/insn.h>
+#include <asm/sections.h>
 
 #define	AARCH64_INSN_IMM_MOVNZ		AARCH64_INSN_IMM_MAX
 #define	AARCH64_INSN_IMM_MOVK		AARCH64_INSN_IMM_16
@@ -394,3 +395,20 @@ int apply_relocate_add(Elf64_Shdr *sechdrs,
 	       me->name, (int)ELF64_R_TYPE(rel[i].r_info), val);
 	return -ENOEXEC;
 }
+
+int module_finalize(const Elf_Ehdr *hdr,
+		    const Elf_Shdr *sechdrs,
+		    struct module *me)
+{
+	const Elf_Shdr *s, *se;
+	const char *secstrs = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
+
+	for (s = sechdrs, se = sechdrs + hdr->e_shnum; s < se; s++) {
+		if (strcmp(".altinstructions", secstrs + s->sh_name) == 0) {
+			apply_alternatives((void *)s->sh_addr, s->sh_size);
+			return 0;
+		}
+	}
+
+	return 0;
+}

arch/arm64/kernel/smp.c

@@ -310,7 +310,7 @@ void cpu_die(void)
 void __init smp_cpus_done(unsigned int max_cpus)
 {
 	pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
-	apply_alternatives();
+	apply_alternatives_all();
 }
 
 void __init smp_prepare_boot_cpu(void)