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.

It is much more clean to use CPUID instruction in usermode to retrieve
this information than to pass AT_HWCAP aux vector from kernel, on
x86.  Still, the change does allow for use of AT_HWCAP on arches where it is
needed, by passing aux array to ifunc_init() initializer which should
prepare arguments for ifunc resolvers.

Current signature for resolvers on x86 is
	func_t iresolve(uint32_t cpu_feature, uint32_t cpu_feature2,
	    uint32_t cpu_stdext_feature, uint32_t cpu_stdext_feature2);
where arguments have identical meaning as the kernel variables of the
same name.  The ABIs allow to use resolvers with the void or shortened
list of arguments.

Reviewed by:	jhb
Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D8448

libexec/rtld-elf/aarch64/reloc.c

@@ -299,6 +299,11 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *defobj,
 	return target;
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT] __unused)
+{
+}
+
 /*
  * Process non-PLT relocations
  */

libexec/rtld-elf/aarch64/rtld_machdep.h

@@ -61,7 +61,10 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 #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(16, align)

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[static 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)
 {

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) \

libexec/rtld-elf/arm/reloc.c

@@ -479,6 +479,11 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *defobj,
 	return target;
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT] __unused)
+{
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {

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;

libexec/rtld-elf/i386/reloc.c

@@ -33,6 +33,7 @@
 
 #include <sys/param.h>
 #include <sys/mman.h>
+#include <machine/cpufunc.h>
 #include <machine/segments.h>
 #include <machine/sysarch.h>
 
@@ -359,7 +360,7 @@ 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 +405,45 @@ 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[static 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;
+
+	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)
 {

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) \

libexec/rtld-elf/mips/reloc.c

@@ -617,6 +617,11 @@ reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *defobj,
 	return target;
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT] __unused)
+{
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {

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;

libexec/rtld-elf/powerpc/reloc.c

@@ -619,6 +619,11 @@ init_pltgot(Obj_Entry *obj)
 	 */
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT] __unused)
+{
+}
+
 void
 allocate_initial_tls(Obj_Entry *list)
 {

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.
  */

libexec/rtld-elf/powerpc64/reloc.c

@@ -523,6 +523,11 @@ init_pltgot(Obj_Entry *obj)
 #endif
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT] __unused)
+{
+}
+
 void
 allocate_initial_tls(Obj_Entry *list)
 {

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.
  */

libexec/rtld-elf/riscv/reloc.c

@@ -366,6 +366,11 @@ reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
 	return (0);
 }
 
+void
+ifunc_init(Elf_Auxinfo aux_info[static AT_COUNT] __unused)
+{
+}
+
 void
 allocate_initial_tls(Obj_Entry *objs)
 {

libexec/rtld-elf/riscv/rtld_machdep.h

@@ -78,6 +78,9 @@ Elf_Addr reloc_jmpslot(Elf_Addr *where, Elf_Addr target,
 	__asm __volatile("mv    gp, %0" :: "r"(old1));			\
 })
 
+#define	call_ifunc_resolver(ptr) \
+	(((Elf_Addr (*)(void))ptr)())
+
 /*
  * Lazy binding entry point, called via PLT.
  */

libexec/rtld-elf/rtld.c

@@ -642,6 +642,7 @@ _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,
@@ -690,7 +691,7 @@ 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);
 }
 

libexec/rtld-elf/rtld.h

@@ -367,6 +367,7 @@ 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[static AT_COUNT]);
 void init_pltgot(Obj_Entry *);
 void lockdflt_init(void);
 void digest_notes(Obj_Entry *, Elf_Addr, Elf_Addr);

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[static AT_COUNT] __unused)
+{
+}
+
 /*
  * Install rtld function call into this PLT slot.
  */

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)