Commit efb734d8 authored by Chris Metcalf's avatar Chris Metcalf

arch/tile: make sure to build memcpy_user_64 without frame pointer

Add a comment explaining why this is important, and add a CFLAGS_REMOVE
clause to the Makefile to make sure it happens.
Signed-off-by: default avatarChris Metcalf <cmetcalf@tilera.com>
parent b287f696
...@@ -7,6 +7,7 @@ lib-y = cacheflush.o checksum.o cpumask.o delay.o uaccess.o \ ...@@ -7,6 +7,7 @@ lib-y = cacheflush.o checksum.o cpumask.o delay.o uaccess.o \
strchr_$(BITS).o strlen_$(BITS).o strchr_$(BITS).o strlen_$(BITS).o
ifeq ($(CONFIG_TILEGX),y) ifeq ($(CONFIG_TILEGX),y)
CFLAGS_REMOVE_memcpy_user_64.o = -fno-omit-frame-pointer
lib-y += memcpy_user_64.o lib-y += memcpy_user_64.o
else else
lib-y += atomic_32.o atomic_asm_32.o memcpy_tile64.o lib-y += atomic_32.o atomic_asm_32.o memcpy_tile64.o
......
...@@ -14,7 +14,13 @@ ...@@ -14,7 +14,13 @@
* Do memcpy(), but trap and return "n" when a load or store faults. * Do memcpy(), but trap and return "n" when a load or store faults.
* *
* Note: this idiom only works when memcpy() compiles to a leaf function. * Note: this idiom only works when memcpy() compiles to a leaf function.
* If "sp" is updated during memcpy, the "jrp lr" will be incorrect. * Here leaf function not only means it does not have calls, but also
* requires no stack operations (sp, stack frame pointer) and no
* use of callee-saved registers, else "jrp lr" will be incorrect since
* unwinding stack frame is bypassed. Since memcpy() is not complex so
* these conditions are satisfied here, but we need to be careful when
* modifying this file. This is not a clean solution but is the best
* one so far.
* *
* Also note that we are capturing "n" from the containing scope here. * Also note that we are capturing "n" from the containing scope here.
*/ */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment