| Message ID | c25a3784dd29348b418904777d3bc5163342b884.1484866680.git.segher@kernel.crashing.org |
|---|---|
| State | New |
| Headers | show |
On Thu, Jan 19, 2017 at 6:00 PM, Segher Boessenkool <segher@kernel.crashing.org> wrote: > I foolishly tested this with r241087 reverted. After that revision > default_stack_protect_guard is no longer called if the compiler defaults > to using the TLS guard, which of course is the wrong thing to do if > there is some other way to enable the global guard. > > This fixes it. > > Is this okay for trunk? Tested on powerpc64-linux {-m32,-m64}; also > will test on powerpc64le-linux. > > > Segher > > > 2017-01-19 Segher Boessenkool <segher@kernel.crashing.org> > > PR target/79140 > * rs6000.c (TARGET_STACK_PROTECT_GUARD): Unconditionally define to > rs6000_init_stack_protect_guard. > (rs6000_init_stack_protect_guard): New function. Okay. Thanks, David
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c index 230e52a..db1145d 100644 --- a/gcc/config/rs6000/rs6000.c +++ b/gcc/config/rs6000/rs6000.c @@ -1759,10 +1759,8 @@ static const struct attribute_spec rs6000_attribute_table[] = #define TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION \ rs6000_builtin_md_vectorized_function -#ifdef TARGET_THREAD_SSP_OFFSET #undef TARGET_STACK_PROTECT_GUARD -#define TARGET_STACK_PROTECT_GUARD hook_tree_void_null -#endif +#define TARGET_STACK_PROTECT_GUARD rs6000_init_stack_protect_guard #if !TARGET_MACHO #undef TARGET_STACK_PROTECT_FAIL @@ -9480,6 +9478,17 @@ rs6000_legitimize_tls_address (rtx addr, enum tls_model model) return dest; } +/* Only create the global variable for the stack protect guard if we are using + the global flavor of that guard. */ +static tree +rs6000_init_stack_protect_guard (void) +{ + if (rs6000_stack_protector_guard == SSP_GLOBAL) + return default_stack_protect_guard (); + + return NULL_TREE; +} + /* Implement TARGET_CANNOT_FORCE_CONST_MEM. */ static bool
I foolishly tested this with r241087 reverted. After that revision default_stack_protect_guard is no longer called if the compiler defaults to using the TLS guard, which of course is the wrong thing to do if there is some other way to enable the global guard. This fixes it. Is this okay for trunk? Tested on powerpc64-linux {-m32,-m64}; also will test on powerpc64le-linux. Segher 2017-01-19 Segher Boessenkool <segher@kernel.crashing.org> PR target/79140 * rs6000.c (TARGET_STACK_PROTECT_GUARD): Unconditionally define to rs6000_init_stack_protect_guard. (rs6000_init_stack_protect_guard): New function. --- gcc/config/rs6000/rs6000.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-)