diff --git a/arch/tile/Kconfig b/arch/tile/Kconfig
index de599c14c4f23460c8f9d614f335798027932bb2..b2be425244835db5d2f989a7297645f10cb07b0c 100644
--- a/arch/tile/Kconfig
+++ b/arch/tile/Kconfig
@@ -26,6 +26,7 @@ config TILE
 	select HAVE_SYSCALL_TRACEPOINTS
 	select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
 	select HAVE_DEBUG_STACKOVERFLOW
+	select ARCH_WANT_FRAME_POINTERS
 
 # FIXME: investigate whether we need/want these options.
 #	select HAVE_IOREMAP_PROT
diff --git a/arch/tile/lib/Makefile b/arch/tile/lib/Makefile
index 5d844374b2b18e6aab7d55ac6738ab08781d1c8c..9adfd76fbdd8452134ee077180c82ab3ad4be9ae 100644
--- a/arch/tile/lib/Makefile
+++ b/arch/tile/lib/Makefile
@@ -6,13 +6,13 @@ lib-y = cacheflush.o checksum.o cpumask.o delay.o uaccess.o \
 	memmove.o memcpy_$(BITS).o memchr_$(BITS).o memset_$(BITS).o \
 	strchr_$(BITS).o strlen_$(BITS).o strnlen_$(BITS).o
 
-ifeq ($(CONFIG_TILEGX),y)
-CFLAGS_REMOVE_memcpy_user_64.o = -fno-omit-frame-pointer
-lib-y += memcpy_user_64.o
-else
-lib-y += atomic_32.o atomic_asm_32.o memcpy_tile64.o
-endif
-
+lib-$(CONFIG_TILEGX) += memcpy_user_64.o
+lib-$(CONFIG_TILEPRO) += atomic_32.o atomic_asm_32.o memcpy_tile64.o
 lib-$(CONFIG_SMP) += spinlock_$(BITS).o usercopy_$(BITS).o
 
 obj-$(CONFIG_MODULES) += exports.o
+
+# The finv_buffer_remote() and copy_{to,from}_user() routines can't
+# have -pg added, since they both rely on being leaf functions.
+CFLAGS_REMOVE_cacheflush.o = -pg
+CFLAGS_REMOVE_memcpy_user_64.o = -pg
diff --git a/arch/tile/lib/cacheflush.c b/arch/tile/lib/cacheflush.c
index 2238b40abf3c9c233938270616c2bb953a035ae4..9c0ec22009a5ed833fab76a6e8ee94a9c1a1685f 100644
--- a/arch/tile/lib/cacheflush.c
+++ b/arch/tile/lib/cacheflush.c
@@ -36,7 +36,8 @@ static inline void force_load(char *p)
  * core (if "!hfh") or homed via hash-for-home (if "hfh"), waiting
  * until the memory controller holds the flushed values.
  */
-void finv_buffer_remote(void *buffer, size_t size, int hfh)
+void __attribute__((optimize("omit-frame-pointer")))
+finv_buffer_remote(void *buffer, size_t size, int hfh)
 {
 	char *p, *base;
 	size_t step_size, load_count;
diff --git a/arch/tile/lib/memcpy_64.c b/arch/tile/lib/memcpy_64.c
index 46fc1600c17d158263819e612ff0a1ff2ee61741..4815354b8cd2c0bc50192f43b568b48bc92ba47e 100644
--- a/arch/tile/lib/memcpy_64.c
+++ b/arch/tile/lib/memcpy_64.c
@@ -54,7 +54,8 @@ void *memcpy(void *__restrict dstv, const void *__restrict srcv, size_t n)
  * macros to return a count of uncopied bytes due to mm fault.
  */
 #define RETVAL 0
-int USERCOPY_FUNC(void *__restrict dstv, const void *__restrict srcv, size_t n)
+int __attribute__((optimize("omit-frame-pointer")))
+USERCOPY_FUNC(void *__restrict dstv, const void *__restrict srcv, size_t n)
 #endif
 {
 	char *__restrict dst1 = (char *)dstv;