Yury Khrustalev <yury.khrusta...@arm.com> writes:
> From: Richard Ball <richard.b...@arm.com>
>
> This patch adds a new testcase and docs
> for the indirect_return attribute.
>
> gcc/ChangeLog:
>
> * doc/extend.texi: Add AArch64 docs for indirect_return
> attribute.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/aarch64/indirect_return.c: New test.
> Co-authored-by: Yury Khrustalev <yury.khrusta...@arm.com>
> ---
> gcc/doc/extend.texi | 5 ++++
> .../gcc.target/aarch64/indirect_return.c | 25 +++++++++++++++++++
> 2 files changed, 30 insertions(+)
> create mode 100644 gcc/testsuite/gcc.target/aarch64/indirect_return.c
>
> diff --git a/gcc/doc/extend.texi b/gcc/doc/extend.texi
> index 42bd567119d..45e2b3ec569 100644
> --- a/gcc/doc/extend.texi
> +++ b/gcc/doc/extend.texi
> @@ -4760,6 +4760,11 @@ Enable or disable calls to out-of-line helpers to
> implement atomic operations.
> This corresponds to the behavior of the command-line options
> @option{-moutline-atomics} and @option{-mno-outline-atomics}.
>
> +@cindex @code{indirect_return} function attribute, AArch64
> +@item indirect_return
> +Used to inform the compiler that a function may return via
> +an indirect return. Adds a BTI J instruction under
> @option{mbranch-protection=} bti.
"return via an indirect return" doesn't really add much information,
especially since the "indirect" might sound related to "indirect
branch", and all returns are indirect in that sense.
How about going with a variation of the x86 documentation:
The @code{indirect_return} attribute can be applied to a function type
to indicate that the function may return via an indirect branch instead
of via a normal return instruction. For example, this can be true of
functions that implement manual context switching between user space
threads, such as POSIX's @code{swapcontext} function.
> @end table
>
> The above target attributes can be specified as follows:
> diff --git a/gcc/testsuite/gcc.target/aarch64/indirect_return.c
> b/gcc/testsuite/gcc.target/aarch64/indirect_return.c
> new file mode 100644
> index 00000000000..f1ef56d5557
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/indirect_return.c
> @@ -0,0 +1,25 @@
> +/* { dg-do compile } */
> +/* { dg-options "-mbranch-protection=bti" } */
> +
> +int __attribute((indirect_return))
> +foo (int a)
> +{
> + return a;
> +}
> +
> +/*
> +**func1:
> +** hint 34 // bti c
> +** ...
> +** bl foo
> +** hint 36 // bti j
> +** ...
> +** ret
> +*/
> +int
> +func1 (int a, int b)
> +{
> + return foo (a + b);
> +}
> +
> +/* { dg-final { check-function-bodies "**" "" "" } } */
I think we should also check the case of a sibling call from an
indirect_return function to an indirect_return function,
since patch 19/22 specifically optimises that case.
Thanks,
Richard
