On 09/12/2020 17:21, Christoph Müllner wrote: > From: Christoph Muellner <christoph.muell...@theobroma-systems.com> > > It is possible to call aarch64_declare_function_name() and > have cfun not set. Let's sanitize the access to this variable. > > gcc/ > > * config/aarch64/aarch64.c (aarch64_declare_function_name): > Santize access to cfun. > --- > gcc/config/aarch64/aarch64.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c > index 67ffba02d3e..264ccb8beb2 100644 > --- a/gcc/config/aarch64/aarch64.c > +++ b/gcc/config/aarch64/aarch64.c > @@ -19317,7 +19317,8 @@ aarch64_declare_function_name (FILE *stream, const > char* name, > ASM_OUTPUT_TYPE_DIRECTIVE (stream, name, "function"); > ASM_OUTPUT_LABEL (stream, name); > > - cfun->machine->label_is_assembled = true; > + if (cfun) > + cfun->machine->label_is_assembled = true; > } > > /* Implement PRINT_PATCHABLE_FUNCTION_ENTRY. Check if the patch area is > after >
Do you have a simple testcase that demonstrates this? If so, we likely have a regression here that will not only need fixing, but back-porting as well. R.
