amd64 rtld: implement support for TLSDESC relocation

Discussed with:	jrtc27
Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D59269

libexec/rtld-elf/amd64/reloc.c

@@ -53,6 +53,22 @@
 #include "rtld.h"
 #include "rtld_tls.h"
 
+void
+md_obj_entry_init(Obj_Entry *obj)
+{
+	STAILQ_INIT(&obj->tlsdesc_dynargs);
+}
+
+void
+md_obj_entry_fini(Obj_Entry *obj)
+{
+	struct tlsdesc_dynarg *tda, *t;
+
+	STAILQ_FOREACH_SAFE(tda, &obj->tlsdesc_dynargs, link, t) {
+		free(tda);
+	}
+}
+
 /*
  * Process the special R_X86_64_COPY relocations in the main program.  These
  * copy data from a shared object into a region in the main program's BSS
@@ -127,6 +143,61 @@ init_pltgot(Obj_Entry *obj)
 	}
 }
 
+static struct tlsdesc_dynarg *
+reloc_tlsdesc_alloc(Obj_Entry *obj, Elf_Addr tlsoffs)
+{
+	struct tlsdesc_dynarg *tda;
+
+	tda = xmalloc(sizeof(struct tlsdesc_dynarg));
+	tda->tlsinfo.ti_module = obj->tlsindex;
+	tda->tlsinfo.ti_offset = tlsoffs;
+	STAILQ_INSERT_TAIL(&obj->tlsdesc_dynargs, tda, link);
+	return (tda);
+}
+
+static void
+reloc_tlsdesc(const Obj_Entry *obj, const Elf_Rela *rela,
+    struct tlsdesc *where, int flags, RtldLockState *lockstate)
+{
+	const Elf_Sym *def;
+	const Obj_Entry *defobj;
+	Elf_Addr offs;
+
+	dbg("reloc_tlsdesc obj %s rela %p where %p", obj->path, rela, where);
+	offs = 0;
+	if (ELF_R_SYM(rela->r_info) != 0) {
+		def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, flags,
+		    NULL, lockstate);
+		if (def == NULL)
+			rtld_die();
+		if (ELF_ST_TYPE(def->st_info) == STT_GNU_IFUNC) {
+			_rtld_error("%s: IFUNC for TLSDESC reloc",
+			    obj->path);
+			rtld_die();
+		}
+		offs = def->st_value;
+		obj = defobj;
+		if (def->st_shndx == SHN_UNDEF) {
+			/* Weak undefined thread variable */
+			where->entry = rtld_tlsdesc_undef;
+			where->addend = rela->r_addend;
+			return;
+		}
+	}
+	offs += rela->r_addend;
+
+	if (obj->tlsoffset != 0) {
+		/* Variable is in initialy allocated TLS segment */
+		where->entry = rtld_tlsdesc_static;
+		where->offset = offs - obj->tlsoffset;
+	} else {
+		/* TLS offest is unknown at load time, use dynamic resolving */
+		where->entry = rtld_tlsdesc_dynamic;
+		where->arg = reloc_tlsdesc_alloc(__DECONST(Obj_Entry *, obj),
+		    offs);
+	}
+}
+
 /* Process the non-PLT relocations. */
 int
 reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
@@ -313,6 +384,10 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
 		case R_X86_64_IRELATIVE:
 			obj->irelative_nonplt = true;
 			break;
+		case R_X86_64_TLSDESC:
+			reloc_tlsdesc(obj, rela, (struct tlsdesc *)where,
+			    flags, lockstate);
+			break;
 
 		/*
 		 * missing:
@@ -334,7 +409,7 @@ done:
 
 /* Process the PLT relocations. */
 int
-reloc_plt(Obj_Entry *obj, int flags __unused, RtldLockState *lockstate __unused)
+reloc_plt(Obj_Entry *obj, int flags, RtldLockState *lockstate)
 {
 	const Elf_Rela *relalim;
 	const Elf_Rela *rela;
@@ -344,10 +419,10 @@ reloc_plt(Obj_Entry *obj, int flags __unused, RtldLockState *lockstate __unused)
 	for (rela = obj->pltrela; rela < relalim; rela++) {
 		Elf_Addr *where;
 
+		where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 		switch (ELF_R_TYPE(rela->r_info)) {
 		case R_X86_64_JMP_SLOT:
 			/* Relocate the GOT slot pointing into the PLT. */
-			where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 			*where += (Elf_Addr)obj->relocbase;
 			break;
 
@@ -355,6 +430,11 @@ reloc_plt(Obj_Entry *obj, int flags __unused, RtldLockState *lockstate __unused)
 			obj->irelative = true;
 			break;
 
+		case R_X86_64_TLSDESC:
+			reloc_tlsdesc(obj, rela, (struct tlsdesc *)where,
+			    SYMLOOK_IN_PLT | flags, lockstate);
+			break;
+
 		default:
 			_rtld_error("Unknown relocation type %x in PLT",
 			    (unsigned int)ELF_R_TYPE(rela->r_info));
@@ -580,3 +660,13 @@ calculate_first_tls_offset(size_t size, size_t align, size_t offset)
 {
 	return (calculate_tls_offset(0, 0, size, align, offset));
 }
+
+ptrdiff_t
+rtld_tlsdesc_dynamic_impl(struct tlsdesc_dynarg *tda, struct tcb *tcb)
+{
+	tls_index *ti;
+
+	ti = &tda->tlsinfo;
+	return ((ptrdiff_t)tls_get_addr_common(tcb, ti->ti_module,
+	    ti->ti_offset) - (ptrdiff_t)tcb);
+}

libexec/rtld-elf/amd64/rtld_machdep.h

@@ -35,10 +35,14 @@
 
 struct Struct_Obj_Entry;
 
-#define	MD_OBJ_ENTRY
+#define	MD_OBJ_ENTRY			\
+	STAILQ_HEAD(, tlsdesc_dynarg) tlsdesc_dynargs;
 
-#define	MD_OBJ_ENTRY_INIT(obj)
-#define	MD_OBJ_ENTRY_FINI(obj)
+#define	MD_OBJ_ENTRY_INIT(obj)	md_obj_entry_init(obj)
+#define	MD_OBJ_ENTRY_FINI(obj)	md_obj_entry_fini(obj)
+
+void md_obj_entry_init(struct Struct_Obj_Entry *obj);
+void md_obj_entry_fini(struct Struct_Obj_Entry *obj);
 
 /* Return the address of the .dynamic section in the dynamic linker. */
 Elf_Dyn *rtld_dynamic_addr(void);
@@ -80,9 +84,35 @@ typedef struct {
 
 void *__tls_get_addr(tls_index *ti) __exported;
 
+#define	__NONE_ABI	/* Cannot be expressed in C */
+
+struct tlsdesc_dynarg
+{
+	tls_index tlsinfo;
+	STAILQ_ENTRY(tlsdesc_dynarg) link;
+};
+
+struct tlsdesc
+{
+	ptrdiff_t (*entry)(struct tlsdesc *) __NONE_ABI;
+	union {
+		Elf_Ssize addend;
+		Elf_Size offset;
+		struct tlsdesc_dynarg *arg;
+	};
+};
+
 #define md_abi_variant_hook(x)
 
 size_t calculate_first_tls_offset(size_t size, size_t align, size_t offset);
 size_t calculate_tls_offset(size_t prev_offset, size_t prev_size, size_t size,
     size_t align, size_t offset);
+
+ptrdiff_t rtld_tlsdesc_dynamic_impl(struct tlsdesc_dynarg *tda,
+    struct tcb *tcb);
+
+ptrdiff_t rtld_tlsdesc_static(struct tlsdesc *) __NONE_ABI;
+ptrdiff_t rtld_tlsdesc_undef(struct tlsdesc *) __NONE_ABI;
+ptrdiff_t rtld_tlsdesc_dynamic(struct tlsdesc *) __NONE_ABI;
+
 #endif

libexec/rtld-elf/amd64/rtld_start.S

@@ -184,15 +184,55 @@ rtld_tlsdesc_undef:
 	.globl	rtld_tlsdesc_undef
 	.type	rtld_tlsdesc_undef,@function
 	.cfi_startproc
+	movq	8(%rax), %rax
+	subq	%fs:0, %rax
 	ret
 	.cfi_endproc
 	.size	rtld_tlsdesc_undef, . - rtld_tlsdesc_undef
 
 rtld_tlsdesc_dynamic:
-	.globl	rtld_tlsdesc_static
+	.globl	rtld_tlsdesc_dynamic
 	.type	rtld_tlsdesc_dynamic,@function
 	.cfi_startproc
+	pushq	%rbp
+	/* XXXKIB cfi */
+	movq	%rsp,%rbp
+	pushq	%rdx
+	pushq	%rcx
+	pushq	%rsi
+	pushq	%rdi
+	movq	8(%rax), %rdi			/* arg->arg */
+	movq	%fs:0, %rsi			/* tcb */
+
+	movq	8(%rsi), %rcx			/* dtv */
+	movq	(%rcx), %rdx			/* dtv_gen */
+	cmpq	%rdx, tls_dtv_generation(%rip)
+	jne	2f
+	movq	(%rdi), %rax			/* arg->tlsinfo.ti_module */
+	dec	%rax
+	movq	16(%rcx, %rax, 8), %rax		/* dtv_tls */
+	testq	%rax, %rax
+	je	2f
+	addq	8(%rdi), %rax			/* arg->offset */
+	subq	%rsi, %rax
+1:
+	popq	%rdi
+	popq	%rsi
+	popq	%rcx
+	popq	%rdx
+	popq	%rbp
 	ret
+2:
+	pushq	%r8
+	pushq	%r9
+	pushq	%r10
+	pushq	%r11
+	call	rtld_tlsdesc_dynamic_impl
+	popq	%r11
+	popq	%r10
+	popq	%r9
+	popq	%r8
+	jmp	1b
 	.cfi_endproc
 	.size	rtld_tlsdesc_dynamic, . - rtld_tlsdesc_dynamic