MFC r308689:

Pass CPUID[1] %edx (cpu_feature), %ecx (cpu_feature2) and
CPUID[7].%ebx (cpu_stdext_feature), %ecx (cpu_stdext_feature2) to the
ifunc resolvers on x86.

MFC r308925:
Adjust r308689 to make rtld compilable with either in-tree or
(hopefully) stock gcc 4.2.1 on i386 and other arches.

18 files changed, 142 insertions, 4 deletions

diff --git a/libexec/rtld-elf/amd64/reloc.c b/libexec/rtld-elf/amd64/reloc.c
index 96cf93c..f13789f 100644
--- a/libexec/rtld-elf/amd64/reloc.c
+++ b/libexec/rtld-elf/amd64/reloc.c
@@ -34,6 +34,7 @@
 #include <sys/param.h>
 #include <sys/mman.h>
 #include <machine/sysarch.h>
+#include <machine/cpufunc.h>
 
 #include <dlfcn.h>
 #include <err.h>
@@ -406,7 +407,7 @@ reloc_iresolve(Obj_Entry *obj, RtldLockState *lockstate)
 	  ptr = (Elf_Addr *)(obj->relocbase + rela->r_addend);
 	  where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
 	  lock_release(rtld_bind_lock, lockstate);
-	  target = ((Elf_Addr (*)(void))ptr)();
+	  target = call_ifunc_resolver(ptr);
 	  wlock_acquire(rtld_bind_lock, lockstate);
 	  *where = target;
 	  break;
@@ -450,6 +451,25 @@ reloc_gnu_ifunc(Obj_Entry *obj, int flags, RtldLockState *lockstate)
     return (0);
 }
 
+uint32_t cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2;
+
+void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+	u_int p[4], cpu_high;
+
+	do_cpuid(1, p);
+	cpu_feature = p[3];
+	cpu_feature2 = p[2];
+	do_cpuid(0, p);
+	cpu_high = p[0];
+	if (cpu_high >= 7) {
+		cpuid_count(7, 0, p);
+		cpu_stdext_feature = p[1];
+		cpu_stdext_feature2 = p[2];
+	}
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {
diff --git a/libexec/rtld-elf/amd64/rtld_machdep.h b/libexec/rtld-elf/amd64/rtld_machdep.h
index cb5e9a1..d6d7dff 100644
--- a/libexec/rtld-elf/amd64/rtld_machdep.h
+++ b/libexec/rtld-elf/amd64/rtld_machdep.h
@@ -61,6 +61,14 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+extern uint32_t cpu_feature;
+extern uint32_t cpu_feature2;
+extern uint32_t cpu_stdext_feature;
+extern uint32_t cpu_stdext_feature2;
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(uint32_t, uint32_t, uint32_t, uint32_t))ptr)( \
+	    cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2))
+
 #define round(size, align) \
 	(((size) + (align) - 1) & ~((align) - 1))
 #define calculate_first_tls_offset(size, align) \
diff --git a/libexec/rtld-elf/arm/reloc.c b/libexec/rtld-elf/arm/reloc.c
index 8b99514..40e6d7c 100644
--- a/libexec/rtld-elf/arm/reloc.c
+++ b/libexec/rtld-elf/arm/reloc.c
@@ -435,6 +435,11 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *defobj,
 }
 
 void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+}
+
+void
 allocate_initial_tls(Obj_Entry *objs)
 {
 #ifdef ARM_TP_ADDRESS
diff --git a/libexec/rtld-elf/arm/rtld_machdep.h b/libexec/rtld-elf/arm/rtld_machdep.h
index f980de0..2c6eb31 100644
--- a/libexec/rtld-elf/arm/rtld_machdep.h
+++ b/libexec/rtld-elf/arm/rtld_machdep.h
@@ -51,6 +51,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(void))ptr)())
+
 #define	TLS_TCB_SIZE	8
 typedef struct {
 	unsigned long ti_module;
diff --git a/libexec/rtld-elf/i386/reloc.c b/libexec/rtld-elf/i386/reloc.c
index 2d6021c..42efcf7 100644
--- a/libexec/rtld-elf/i386/reloc.c
+++ b/libexec/rtld-elf/i386/reloc.c
@@ -359,7 +359,7 @@ reloc_iresolve(Obj_Entry *obj, RtldLockState *lockstate)
 	case R_386_IRELATIVE:
 	  where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
 	  lock_release(rtld_bind_lock, lockstate);
-	  target = ((Elf_Addr (*)(void))(obj->relocbase + *where))();
+	  target = call_ifunc_resolver(obj->relocbase + *where);
 	  wlock_acquire(rtld_bind_lock, lockstate);
 	  *where = target;
 	  break;
@@ -404,6 +404,58 @@ reloc_gnu_ifunc(Obj_Entry *obj, int flags, RtldLockState *lockstate)
     return (0);
 }
 
+uint32_t cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2;
+
+static void
+rtld_cpuid_count(int idx, int cnt, u_int *p)
+{
+
+	__asm __volatile(
+	    "	pushl	%%ebx\n"
+	    "	cpuid\n"
+	    "	movl	%%ebx,%1\n"
+	    "	popl	%%ebx\n"
+	    : "=a" (p[0]), "=r" (p[1]), "=c" (p[2]), "=d" (p[3])
+	    :  "0" (idx), "2" (cnt));
+}
+
+void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+	u_int p[4], cpu_high;
+	int cpuid_supported;
+
+	__asm __volatile(
+	    "	pushfl\n"
+	    "	popl	%%eax\n"
+	    "	movl    %%eax,%%ecx\n"
+	    "	xorl    $0x200000,%%eax\n"
+	    "	pushl	%%eax\n"
+	    "	popfl\n"
+	    "	pushfl\n"
+	    "	popl    %%eax\n"
+	    "	xorl    %%eax,%%ecx\n"
+	    "	je	1f\n"
+	    "	movl	$1,%0\n"
+	    "	jmp	2f\n"
+	    "1:	movl	$0,%0\n"
+	    "2:\n"
+	    : "=r" (cpuid_supported) : : "eax", "ecx");
+	if (!cpuid_supported)
+		return;
+
+	rtld_cpuid_count(1, 0, p);
+	cpu_feature = p[3];
+	cpu_feature2 = p[2];
+	rtld_cpuid_count(0, 0, p);
+	cpu_high = p[0];
+	if (cpu_high >= 7) {
+		rtld_cpuid_count(7, 0, p);
+		cpu_stdext_feature = p[1];
+		cpu_stdext_feature2 = p[2];
+	}
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {
diff --git a/libexec/rtld-elf/i386/rtld_machdep.h b/libexec/rtld-elf/i386/rtld_machdep.h
index 5c328da..850b170 100644
--- a/libexec/rtld-elf/i386/rtld_machdep.h
+++ b/libexec/rtld-elf/i386/rtld_machdep.h
@@ -61,6 +61,14 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+extern uint32_t cpu_feature;
+extern uint32_t cpu_feature2;
+extern uint32_t cpu_stdext_feature;
+extern uint32_t cpu_stdext_feature2;
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(uint32_t, uint32_t, uint32_t, uint32_t))ptr)( \
+	    cpu_feature, cpu_feature2, cpu_stdext_feature, cpu_stdext_feature2))
+
 #define round(size, align) \
 	(((size) + (align) - 1) & ~((align) - 1))
 #define calculate_first_tls_offset(size, align) \
diff --git a/libexec/rtld-elf/ia64/reloc.c b/libexec/rtld-elf/ia64/reloc.c
index 539b7ef..76f06f5 100644
--- a/libexec/rtld-elf/ia64/reloc.c
+++ b/libexec/rtld-elf/ia64/reloc.c
@@ -603,6 +603,11 @@ call_init_pointer(const Obj_Entry *obj, Elf_Addr target)
 	((InitArrFunc) &fptr)(main_argc, main_argv, environ);
 }
 
+void  
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+}
+
 /* Initialize the special PLT entries. */
 void
 init_pltgot(Obj_Entry *obj)
diff --git a/libexec/rtld-elf/ia64/rtld_machdep.h b/libexec/rtld-elf/ia64/rtld_machdep.h
index b00a90fc..f1e2554 100644
--- a/libexec/rtld-elf/ia64/rtld_machdep.h
+++ b/libexec/rtld-elf/ia64/rtld_machdep.h
@@ -54,6 +54,9 @@ void *make_function_pointer(const Elf_Sym *, const struct Struct_Obj_Entry *);
 void call_initfini_pointer(const struct Struct_Obj_Entry *, Elf_Addr);
 void call_init_pointer(const struct Struct_Obj_Entry *, Elf_Addr);
 
+#define        call_ifunc_resolver(ptr) \
+       (((Elf_Addr (*)(void))ptr)())
+
 #define	TLS_TCB_SIZE	16
 
 #define round(size, align) \
diff --git a/libexec/rtld-elf/mips/reloc.c b/libexec/rtld-elf/mips/reloc.c
index 809adb8..2ca915e 100644
--- a/libexec/rtld-elf/mips/reloc.c
+++ b/libexec/rtld-elf/mips/reloc.c
@@ -618,6 +618,11 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *defobj,
 }
 
 void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+}
+
+void
 allocate_initial_tls(Obj_Entry *objs)
 {
 	char *tls;
diff --git a/libexec/rtld-elf/mips/rtld_machdep.h b/libexec/rtld-elf/mips/rtld_machdep.h
index befbf13..4d2c9d9 100644
--- a/libexec/rtld-elf/mips/rtld_machdep.h
+++ b/libexec/rtld-elf/mips/rtld_machdep.h
@@ -52,6 +52,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(void))ptr)())
+
 typedef struct {
 	unsigned long ti_module;
 	unsigned long ti_offset;
diff --git a/libexec/rtld-elf/powerpc/reloc.c b/libexec/rtld-elf/powerpc/reloc.c
index a9a16de..f78c1d5 100644
--- a/libexec/rtld-elf/powerpc/reloc.c
+++ b/libexec/rtld-elf/powerpc/reloc.c
@@ -620,6 +620,11 @@ init_pltgot(Obj_Entry *obj)
 }
 
 void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+}
+
+void
 allocate_initial_tls(Obj_Entry *list)
 {
 	register Elf_Addr **tp __asm__("r2");
diff --git a/libexec/rtld-elf/powerpc/rtld_machdep.h b/libexec/rtld-elf/powerpc/rtld_machdep.h
index 1ddf1bc..9ab0e75 100644
--- a/libexec/rtld-elf/powerpc/rtld_machdep.h
+++ b/libexec/rtld-elf/powerpc/rtld_machdep.h
@@ -51,6 +51,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(void))ptr)())
+
 /*
  * Lazy binding entry point, called via PLT.
  */
diff --git a/libexec/rtld-elf/powerpc64/reloc.c b/libexec/rtld-elf/powerpc64/reloc.c
index b33f6fe..79fbd2d 100644
--- a/libexec/rtld-elf/powerpc64/reloc.c
+++ b/libexec/rtld-elf/powerpc64/reloc.c
@@ -488,6 +488,11 @@ init_pltgot(Obj_Entry *obj)
 }
 
 void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+}
+
+void
 allocate_initial_tls(Obj_Entry *list)
 {
 	Elf_Addr **tp;
diff --git a/libexec/rtld-elf/powerpc64/rtld_machdep.h b/libexec/rtld-elf/powerpc64/rtld_machdep.h
index b88ed9d..ce2139c 100644
--- a/libexec/rtld-elf/powerpc64/rtld_machdep.h
+++ b/libexec/rtld-elf/powerpc64/rtld_machdep.h
@@ -51,6 +51,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(void))ptr)())
+
 /*
  * Lazy binding entry point, called via PLT.
  */
diff --git a/libexec/rtld-elf/rtld.c b/libexec/rtld-elf/rtld.c
index 0b72f85..75fda31 100644
--- a/libexec/rtld-elf/rtld.c
+++ b/libexec/rtld-elf/rtld.c
@@ -635,6 +635,7 @@ _rtld(Elf_Addr *sp, func_ptr_type *exit_proc, Obj_Entry **objp)
     r_debug_state(NULL, &obj_main->linkmap); /* say hello to gdb! */
 
     map_stacks_exec(NULL);
+    ifunc_init(aux);
 
     dbg("resolving ifuncs");
     if (resolve_objects_ifunc(obj_main,
@@ -683,7 +684,7 @@ rtld_resolve_ifunc(const Obj_Entry *obj, const Elf_Sym *def)
 	Elf_Addr target;
 
 	ptr = (void *)make_function_pointer(def, obj);
-	target = ((Elf_Addr (*)(void))ptr)();
+	target = call_ifunc_resolver(ptr);
 	return ((void *)target);
 }
 
diff --git a/libexec/rtld-elf/rtld.h b/libexec/rtld-elf/rtld.h
index d63222d..cb0448d 100644
--- a/libexec/rtld-elf/rtld.h
+++ b/libexec/rtld-elf/rtld.h
@@ -380,6 +380,7 @@ void dump_Elf_Rela(Obj_Entry *, const Elf_Rela *, u_long);
 unsigned long elf_hash(const char *);
 const Elf_Sym *find_symdef(unsigned long, const Obj_Entry *,
   const Obj_Entry **, int, SymCache *, struct Struct_RtldLockState *);
+void ifunc_init(Elf_Auxinfo *);
 void init_pltgot(Obj_Entry *);
 void lockdflt_init(void);
 void digest_notes(Obj_Entry *, Elf_Addr, Elf_Addr);
diff --git a/libexec/rtld-elf/sparc64/reloc.c b/libexec/rtld-elf/sparc64/reloc.c
index 242fb8e..ab16879 100644
--- a/libexec/rtld-elf/sparc64/reloc.c
+++ b/libexec/rtld-elf/sparc64/reloc.c
@@ -786,6 +786,11 @@ reloc_jmpslot(Elf_Addr *wherep, Elf_Addr target, const Obj_Entry *obj,
 	return (target);
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[__min_size(AT_COUNT)] __unused)
+{
+}
+
 /*
  * Install rtld function call into this PLT slot.
  */
diff --git a/libexec/rtld-elf/sparc64/rtld_machdep.h b/libexec/rtld-elf/sparc64/rtld_machdep.h
index 44fe2cf..45e0202 100644
--- a/libexec/rtld-elf/sparc64/rtld_machdep.h
+++ b/libexec/rtld-elf/sparc64/rtld_machdep.h
@@ -53,7 +53,10 @@ Elf_Addr reloc_jmpslot(Elf_Addr *, Elf_Addr,
 #define call_init_pointer(obj, target) \
 	(((InitArrFunc)(target))(main_argc, main_argv, environ))
 
-#define round(size, align) \
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(void))ptr)())
+
+#define round(size, align)				\
 	(((size) + (align) - 1) & ~((align) - 1))
 #define calculate_first_tls_offset(size, align) \
 	round(size, align)