xtensa: add FDPIC support

This change implements Xtensa FDPIC ABI as specified in the first
version of the following document:

 https://github.com/jcmvbkbc/xtensa-abi/blob/master/fdpic-xtensa.txt

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>

5 files changed, 253 insertions, 29 deletions

diff --git a/ldso/ldso/xtensa/dl-inlines.h b/ldso/ldso/xtensa/dl-inlines.h
new file mode 100644
index 000000000..8fdf6eb48
--- /dev/null
+++ b/ldso/ldso/xtensa/dl-inlines.h
@@ -0,0 +1 @@
+#include "../fdpic/dl-inlines.h"
diff --git a/ldso/ldso/xtensa/dl-startup.h b/ldso/ldso/xtensa/dl-startup.h
index c9350c0f2..2a453752a 100644
--- a/ldso/ldso/xtensa/dl-startup.h
+++ b/ldso/ldso/xtensa/dl-startup.h
@@ -7,6 +7,68 @@
  * Parts taken from glibc/sysdeps/xtensa/dl-machine.h.
  */
 
+#if defined(__FDPIC__)
+__asm__ (
+    "	.text\n"
+    "	.align  4\n"
+    "   .literal_position\n"
+    "	.global _start\n"
+    "	.type   _start, @function\n"
+    "	.hidden _start\n"
+    "_start:\n"
+    "	.begin	no-transform\n"
+    "	_call0  1f\n"
+    "2:\n"
+    "	.end	no-transform\n"
+    "	.align  4\n"
+    "1:\n"
+#if defined(__XTENSA_CALL0_ABI__)
+    "	movi	a15, 2b\n"
+    "	sub	a15, a0, a15\n"
+
+    /* Save FDPIC pointers in callee-saved registers */
+    "	mov	a12, a4\n"
+    "	mov	a13, a5\n"
+    "	mov	a14, a6\n"
+
+    /* Call __self_reloc */
+    "	mov	a2, a5\n"
+    "	movi	a3, __ROFIXUP_LIST__\n"
+    "	add	a3, a3, a15\n"
+    "	movi	a4, __ROFIXUP_END__\n"
+    "	add	a4, a4, a15\n"
+    "	movi    a0, __self_reloc\n"
+    "	add     a0, a0, a15\n"
+    "	callx0  a0\n"
+
+    /* call _dl_start */
+    "	mov	a3, a12\n"
+    "	mov	a4, a13\n"
+    "	mov	a5, a14\n"
+    "	mov	a7, sp\n"
+    "	addi	sp, sp, -16\n"
+    "	mov	a6, sp\n"
+    "	mov	a11, a2\n"
+    /* a13, interpreter map is no longer needed, save interpreter GOT there */
+    "	mov	a13, a2\n"
+    "	movi	a0, _dl_start\n"
+    "	add	a0, a0, a15\n"
+    "	callx0	a0\n"
+
+    /* call main */
+    "	l32i	a0, sp, 0\n"
+    "	l32i	a11, sp, 4\n"
+    "	addi	sp, sp, 16\n"
+    "	mov	a4, a12\n"
+    "	movi	a5, _dl_fini@GOTOFFFUNCDESC\n"
+    "	add	a5, a5, a13\n"
+    "	mov	a6, a14\n"
+    "	jx	a0\n"
+#else
+#error Unsupported Xtensa ABI
+#endif
+   );
+#else /* __FDPIC__ */
 #ifndef L_rcrt1
 __asm__ (
     "	.text\n"
@@ -83,6 +145,7 @@ __asm__ (
     "	bnez    a6, 3b\n"
     "	j      .Lfixup_stack_ret");
 #endif
+#endif /* __FDPIC__ */
 
 /* Get a pointer to the argv value.  */
 #define GET_ARGV(ARGVP, ARGS) ARGVP = (((unsigned long *) ARGS) + 1)
@@ -90,7 +153,7 @@ __asm__ (
 /* Function calls are not safe until the GOT relocations have been done.  */
 #define NO_FUNCS_BEFORE_BOOTSTRAP
 
-#if defined(__ARCH_USE_MMU__)
+#if defined(__ARCH_USE_MMU__) && !defined(__FDPIC__)
 #define PERFORM_BOOTSTRAP_GOT(tpnt) \
 do { \
 	xtensa_got_location *got_loc; \
@@ -128,3 +191,29 @@ do { \
 	} \
 } while (0)
 #endif
+
+#ifdef __FDPIC__
+#undef DL_START
+#define DL_START(X)   \
+static void  __attribute__ ((used)) \
+_dl_start (Elf32_Addr dl_boot_got_pointer, \
+          struct elf32_fdpic_loadmap *dl_boot_progmap, \
+          struct elf32_fdpic_loadmap *dl_boot_ldsomap, \
+          Elf32_Dyn *dl_boot_ldso_dyn_pointer, \
+          struct funcdesc_value *dl_main_funcdesc, \
+          X)
+
+/*
+ * Transfer control to the user's application, once the dynamic loader
+ * is done.  We return the address of the function's entry point to
+ * _dl_boot, see boot1_arch.h.
+ */
+#define START()	do {							\
+  struct elf_resolve *exec_mod = _dl_loaded_modules;			\
+  dl_main_funcdesc->entry_point = _dl_elf_main;				\
+  while (exec_mod->libtype != elf_executable)				\
+    exec_mod = exec_mod->next;						\
+  dl_main_funcdesc->got_value = exec_mod->loadaddr.got_value;		\
+  return;								\
+} while (0)
+#endif /* __FDPIC__ */
diff --git a/ldso/ldso/xtensa/dl-sysdep.h b/ldso/ldso/xtensa/dl-sysdep.h
index 6b908989a..5aa3e177f 100644
--- a/ldso/ldso/xtensa/dl-sysdep.h
+++ b/ldso/ldso/xtensa/dl-sysdep.h
@@ -26,6 +26,7 @@
    in l_info array.  */
 #define DT_XTENSA(x) (DT_XTENSA_##x - DT_LOPROC + DT_NUM + OS_NUM)
 
+#ifndef __FDPIC__
 typedef struct xtensa_got_location_struct {
   Elf32_Off offset;
   Elf32_Word length;
@@ -86,6 +87,7 @@ typedef struct xtensa_got_location_struct {
     else if (dpnt->d_tag == DT_XTENSA_GOT_LOC_SZ)			\
       dynamic[DT_XTENSA (GOT_LOC_SZ)] = dpnt->d_un.d_val;		\
   } while (0)
+#endif
 
 /* Here we define the magic numbers that this dynamic loader should accept. */
 #define MAGIC1 EM_XTENSA
@@ -115,10 +117,41 @@ elf_machine_dynamic (void)
   return (Elf32_Addr) &_DYNAMIC;
 }
 
+#ifdef __FDPIC__
+
+#define DL_CHECK_LIB_TYPE(epnt, piclib, _dl_progname, libname) \
+do \
+{ \
+  (piclib) = 2; \
+} \
+while (0)
+
+/* We must force strings used early in the bootstrap into the data
+   segment.  */
+#undef SEND_EARLY_STDERR
+#define SEND_EARLY_STDERR(S) \
+  do { /* FIXME: implement */; } while (0)
+
+#undef INIT_GOT
+#include "../fdpic/dl-sysdep.h"
+#undef INIT_GOT
+#define INIT_GOT(GOT_BASE,MODULE) \
+{				\
+  (MODULE)->loadaddr.got_value = (GOT_BASE); \
+  GOT_BASE[0] = ((unsigned long *)&_dl_linux_resolve)[0]; \
+  GOT_BASE[1] = ((unsigned long *)&_dl_linux_resolve)[1]; \
+  GOT_BASE[2] = (unsigned long) MODULE; \
+}
+
+#endif /* __FDPIC__ */
+
 /* Return the run-time load address of the shared object.  */
 static __always_inline Elf32_Addr
 elf_machine_load_address (void)
 {
+#ifdef __FDPIC__
+  return 0;
+#else
   Elf32_Addr addr, tmp;
 
   /* At this point, the runtime linker is being bootstrapped and the GOT
@@ -135,17 +168,48 @@ elf_machine_load_address (void)
 	   : "=a" (addr), "=a" (tmp));
 
   return addr - 3;
+#endif
 }
 
+#ifdef __FDPIC__
+
+/* Need bootstrap relocations */
+#define ARCH_NEEDS_BOOTSTRAP_RELOCS
+
+#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD,SYMTAB)	\
+	switch (ELF_R_TYPE((RELP)->r_info)){			\
+	case R_XTENSA_SYM32:					\
+		*(REL)  = (SYMBOL) + (RELP)->r_addend;		\
+		break;						\
+	case R_XTENSA_RELATIVE:					\
+	case R_XTENSA_NONE:					\
+	default:						\
+		break;						\
+	}
+
 static __always_inline void
-elf_machine_relative (Elf32_Addr load_off, const Elf32_Addr rel_addr,
+elf_machine_relative (DL_LOADADDR_TYPE load_off, const Elf32_Addr rel_addr,
 		      Elf32_Word relative_count)
 {
   Elf32_Rela *rpnt = (Elf32_Rela *) rel_addr;
   while (relative_count--)
     {
+      Elf32_Addr *const reloc_addr = (Elf32_Addr *) DL_RELOC_ADDR(load_off, rpnt->r_offset);
+      *reloc_addr = DL_RELOC_ADDR(load_off, *reloc_addr);
+      rpnt++;
+    }
+}
+#else
+static __always_inline void
+elf_machine_relative (Elf32_Addr load_off, const Elf32_Addr rel_addr,
+                     Elf32_Word relative_count)
+{
+  Elf32_Rela *rpnt = (Elf32_Rela *) rel_addr;
+  while (relative_count--)
+    {
       Elf32_Addr *const reloc_addr = (Elf32_Addr *) (load_off + rpnt->r_offset);
       *reloc_addr += load_off + rpnt->r_addend;
       rpnt++;
     }
 }
+#endif
diff --git a/ldso/ldso/xtensa/dl-tlsdesc.S b/ldso/ldso/xtensa/dl-tlsdesc.S
index 426f2180b..1a8eacff2 100644
--- a/ldso/ldso/xtensa/dl-tlsdesc.S
+++ b/ldso/ldso/xtensa/dl-tlsdesc.S
@@ -24,6 +24,9 @@
 
 	.text
 HIDDEN_ENTRY (_dl_tlsdesc_return)
+#ifdef __FDPIC__
+	l32i		a2, a2, 4
+#endif
 	rur.threadptr	a3
 	add		a2, a2, a3
 	abi_ret
@@ -53,7 +56,9 @@ END (_dl_tlsdesc_return)
 	 */
 
 HIDDEN_ENTRY (_dl_tlsdesc_dynamic)
-
+#ifdef __FDPIC__
+	l32i	a2, a2, 4
+#endif
 	/* dtv_t *dtv = (dtv_t *)THREAD_DTV(); */
 	rur.threadptr	a3
 	l32i	a4, a3, 0
@@ -86,7 +91,8 @@ HIDDEN_ENTRY (_dl_tlsdesc_dynamic)
 #elif defined(__XTENSA_CALL0_ABI__)
 	addi	a1, a1, -16
 	s32i	a0, a1, 0
-	movi	a0, __tls_get_addr
+	movi	a0, JUMPTARGET(__tls_get_addr)
+	FDPIC_LOAD_JUMPTARGET(a0, a11, a0)
 	callx0	a0
 	l32i	a0, a1, 0
 	addi	a1, a1, 16
diff --git a/ldso/ldso/xtensa/elfinterp.c b/ldso/ldso/xtensa/elfinterp.c
index e38a02666..d97f23435 100644
--- a/ldso/ldso/xtensa/elfinterp.c
+++ b/ldso/ldso/xtensa/elfinterp.c
@@ -36,6 +36,13 @@
 #include "tlsdeschtab.h"
 #endif
 
+#ifdef __FDPIC__
+unsigned long
+_dl_linux_resolver (struct elf_resolve *tpnt, int reloc_entry)
+{
+	return 0;
+}
+#else
 unsigned long
 _dl_linux_resolver (struct elf_resolve *tpnt, int reloc_entry)
 {
@@ -83,7 +90,7 @@ _dl_linux_resolver (struct elf_resolve *tpnt, int reloc_entry)
 
 	return (unsigned long) new_addr;
 }
-
+#endif
 
 static int
 _dl_parse (struct elf_resolve *tpnt, struct r_scope_elem *scope,
@@ -145,8 +152,8 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 	int reloc_type;
 	int symtab_index;
 	char *symname;
-#if defined USE_TLS && USE_TLS
-	struct elf_resolve *tls_tpnt = NULL;
+#if defined USE_TLS && USE_TLS || defined (__FDPIC__)
+	struct elf_resolve *def_mod = NULL;
 #endif
 	struct symbol_ref sym_ref;
 	ElfW(Addr) *reloc_addr;
@@ -155,7 +162,7 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 	ElfW(Addr) old_val;
 #endif
 
-	reloc_addr = (ElfW(Addr) *) (tpnt->loadaddr + rpnt->r_offset);
+	reloc_addr = (ElfW(Addr) *) DL_RELOC_ADDR(tpnt->loadaddr, rpnt->r_offset);
 	reloc_type = ELF_R_TYPE (rpnt->r_info);
 	symtab_index = ELF_R_SYM (rpnt->r_info);
 	sym_ref.sym = &symtab[symtab_index];
@@ -164,9 +171,17 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 	symname = strtab + sym_ref.sym->st_name;
 
 	if (symtab_index) {
-		symbol_addr = (ElfW(Addr))
-			_dl_find_hash (symname, scope, tpnt,
-						   elf_machine_type_class (reloc_type), &sym_ref);
+		if (ELF_ST_BIND (sym_ref.sym->st_info) == STB_LOCAL) {
+			symbol_addr = (ElfW(Addr))
+				DL_RELOC_ADDR(tpnt->loadaddr,
+					      symtab[symtab_index].st_value);
+			sym_ref.tpnt = tpnt;
+		} else {
+			symbol_addr = (ElfW(Addr))
+				_dl_find_hash (symname, scope, tpnt,
+					       elf_machine_type_class (reloc_type),
+					       &sym_ref);
+		}
 
 		/*
 		 * We want to allow undefined references to weak symbols - this might
@@ -182,13 +197,13 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 			_dl_debug_lookup (symname, tpnt, &symtab[symtab_index],
 						&sym_ref, elf_machine_type_class(reloc_type));
 		}
-#if defined USE_TLS && USE_TLS
-		tls_tpnt = sym_ref.tpnt;
+#if defined USE_TLS && USE_TLS || defined (__FDPIC__)
+		def_mod = sym_ref.tpnt;
 #endif
 	} else {
 		symbol_addr =symtab[symtab_index].st_value;
-#if defined USE_TLS && USE_TLS
-		tls_tpnt = tpnt;
+#if defined USE_TLS && USE_TLS || defined (__FDPIC__)
+		def_mod = tpnt;
 #endif
 	}
 
@@ -202,6 +217,7 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 
 	case R_XTENSA_GLOB_DAT:
 	case R_XTENSA_JMP_SLOT:
+	case R_XTENSA_SYM32:
 		*reloc_addr = symbol_addr + rpnt->r_addend;
 		break;
 
@@ -219,19 +235,63 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 		break;
 
 	case R_XTENSA_RELATIVE:
-		*reloc_addr += tpnt->loadaddr + rpnt->r_addend;
+		*reloc_addr += DL_RELOC_ADDR(tpnt->loadaddr, rpnt->r_addend);
 		break;
 
+#ifdef __FDPIC__
+	case R_XTENSA_FUNCDESC_VALUE:
+		{
+			struct funcdesc_value *dst = (struct funcdesc_value *) reloc_addr;
+
+			dst->entry_point = (void *) (symbol_addr + rpnt->r_addend);
+			dst->got_value = def_mod->loadaddr.got_value;
+		}
+		break;
+	case R_XTENSA_FUNCDESC:
+		if (symbol_addr)
+			*reloc_addr = (unsigned long)
+				_dl_funcdesc_for((void *) (symbol_addr + rpnt->r_addend),
+						 sym_ref.tpnt->loadaddr.got_value);
+		else
+			/* Relocation against an undefined weak symbol:
+			   set funcdesc to zero.  */
+			*reloc_addr = 0;
+		break;
+	case R_XTENSA_TLS_TPOFF:
+		CHECK_STATIC_TLS((struct link_map *) def_mod);
+		*reloc_addr = symbol_addr + rpnt->r_addend + def_mod->l_tls_offset;
+		break;
+	case R_XTENSA_TLSDESC:
+		{
+			struct tlsdesc *td = (struct tlsdesc *) reloc_addr;
+#ifndef SHARED
+			CHECK_STATIC_TLS((struct link_map *) def_mod);
+#else
+			if (!TRY_STATIC_TLS ((struct link_map *) def_mod))
+			{
+				td->entry = _dl_tlsdesc_dynamic;
+				td->argument = _dl_make_tlsdesc_dynamic((struct link_map *) def_mod,
+									symbol_addr + rpnt->r_addend);
+			}
+			else
+#endif
+			{
+				td->entry = _dl_tlsdesc_return;
+				td->argument = (void *) (symbol_addr + rpnt->r_addend + def_mod->l_tls_offset);
+			}
+		}
+		break;
+#else
 #if defined USE_TLS && USE_TLS
 	case R_XTENSA_TLS_TPOFF:
-		CHECK_STATIC_TLS((struct link_map *) tls_tpnt);
-		*reloc_addr = symbol_addr + tls_tpnt->l_tls_offset + rpnt->r_addend;
+		CHECK_STATIC_TLS((struct link_map *) def_mod);
+		*reloc_addr = symbol_addr + rpnt->r_addend + def_mod->l_tls_offset;
 		break;
 	case R_XTENSA_TLSDESC_FN:
 #ifndef SHARED
-		CHECK_STATIC_TLS((struct link_map *) tls_tpnt);
+		CHECK_STATIC_TLS((struct link_map *) def_mod);
 #else
-		if (!TRY_STATIC_TLS ((struct link_map *) tls_tpnt))
+		if (!TRY_STATIC_TLS ((struct link_map *) def_mod))
 			*reloc_addr = (ElfW(Addr)) _dl_tlsdesc_dynamic;
 		else
 #endif
@@ -239,25 +299,25 @@ _dl_do_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 		break;
 	case R_XTENSA_TLSDESC_ARG:
 #ifndef SHARED
-		CHECK_STATIC_TLS((struct link_map *) tls_tpnt);
+		CHECK_STATIC_TLS((struct link_map *) def_mod);
 #else
-		if (!TRY_STATIC_TLS ((struct link_map *) tls_tpnt))
+		if (!TRY_STATIC_TLS ((struct link_map *) def_mod))
 			*reloc_addr = (ElfW(Addr))
-				_dl_make_tlsdesc_dynamic((struct link_map *) tls_tpnt,
+				_dl_make_tlsdesc_dynamic((struct link_map *) def_mod,
 							 symbol_addr + rpnt->r_addend);
 		else
 #endif
 			*reloc_addr = symbol_addr + rpnt->r_addend +
-				tls_tpnt->l_tls_offset;
+				def_mod->l_tls_offset;
 		break;
 #endif
-
+#endif
 	default:
 		return -1; /* Calls _dl_exit(1).  */
 	}
 #if defined (__SUPPORT_LD_DEBUG__)
 	if (_dl_debug_reloc && _dl_debug_detail)
-		_dl_dprintf (_dl_debug_file, "\tpatched: %x ==> %x @ %x\n",
+		_dl_dprintf (_dl_debug_file, "\tpatched: %x ==> %x @ %p\n",
 					 old_val, *reloc_addr, reloc_addr);
 #endif
 
@@ -275,7 +335,7 @@ _dl_do_lazy_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 	ElfW(Addr) old_val;
 #endif
 
-	reloc_addr = (ElfW(Addr) *) (tpnt->loadaddr + rpnt->r_offset);
+	reloc_addr = (ElfW(Addr) *) DL_RELOC_ADDR(tpnt->loadaddr, rpnt->r_offset);
 	reloc_type = ELF_R_TYPE (rpnt->r_info);
 
 #if defined (__SUPPORT_LD_DEBUG__)
@@ -286,7 +346,7 @@ _dl_do_lazy_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 	case R_XTENSA_JMP_SLOT:
 		/* Perform a RELATIVE reloc on the GOT entry that transfers
 		   to the stub function.  */
-		*reloc_addr += tpnt->loadaddr;
+		*reloc_addr = DL_RELOC_ADDR(tpnt->loadaddr, *reloc_addr);
 		break;
 	case R_XTENSA_NONE:
 		break;
@@ -296,7 +356,7 @@ _dl_do_lazy_reloc (struct elf_resolve *tpnt, struct r_scope_elem *scope,
 
 #if defined (__SUPPORT_LD_DEBUG__)
 	if (_dl_debug_reloc && _dl_debug_detail)
-		_dl_dprintf (_dl_debug_file, "\tpatched: %x ==> %x @ %x\n",
+		_dl_dprintf (_dl_debug_file, "\tpatched: %x ==> %x @ %p\n",
 					 old_val, *reloc_addr, reloc_addr);
 #endif
 	return 0;
@@ -320,3 +380,7 @@ _dl_parse_relocation_information (struct dyn_elf *rpnt,
 	return _dl_parse (rpnt->dyn, scope, rel_addr, rel_size,
 					  _dl_do_reloc);
 }
+
+#ifndef IS_IN_libdl
+# include "../../libc/sysdeps/linux/xtensa/crtreloc.c"
+#endif