[PATCH] RISC-V: properly terminate call chain

[Palmer Dabbelt]

On Thu, 25 Oct 2018 02:53:44 PDT (-0700), schwab@suse.de wrote:
> Mark the ra register as undefined in _start, so that unwinding through
> main works correctly.  Also, don't use a tail call so that ra points after
> the call to __libc_start_main, not after the previous call.
>
> 	[BZ #23125]
> 	* sysdeps/riscv/start.S (ENTRY_POINT): Mark ra as undefined.
> 	Don't use tail call.
> 	* elf/tst-unwind-main.c: New file.
> 	* elf/Makefile (tests): Add tst-unwind-main.
> 	(CFLAGS-tst-unwind-main.c): Define.
> ---
>  elf/Makefile          |  4 +++-
>  elf/tst-unwind-main.c | 38 ++++++++++++++++++++++++++++++++++++++
>  sysdeps/riscv/start.S |  7 ++++++-
>  3 files changed, 47 insertions(+), 2 deletions(-)
>  create mode 100644 elf/tst-unwind-main.c

Thanks.  Your original reply to my patch has been the top of my inbox for about 
6 months now but I keep getting interrupted before I can actually write up the 
patch.

> diff --git a/elf/Makefile b/elf/Makefile
> index 455ec730fc..d277d34b17 100644
> --- a/elf/Makefile
> +++ b/elf/Makefile
> @@ -187,7 +187,7 @@ tests += restest1 preloadtest loadfail multiload origtest resolvfail \
>  	 tst-nodelete2 tst-audit11 tst-audit12 tst-dlsym-error tst-noload \
>  	 tst-latepthread tst-tls-manydynamic tst-nodelete-dlclose \
>  	 tst-debug1 tst-main1 tst-absolute-sym tst-absolute-zero tst-big-note \
> -	 tst-unwind-ctor
> +	 tst-unwind-ctor tst-unwind-main
>  #	 reldep9
>  tests-internal += loadtest unload unload2 circleload1 \
>  	 neededtest neededtest2 neededtest3 neededtest4 \
> @@ -1495,3 +1495,5 @@ $(objpfx)tst-libc_dlvsym-static.out: $(objpfx)tst-libc_dlvsym-dso.so
>  $(objpfx)tst-big-note: $(objpfx)tst-big-note-lib.so
>
>  $(objpfx)tst-unwind-ctor: $(objpfx)tst-unwind-ctor-lib.so
> +
> +CFLAGS-tst-unwind-main.c += -funwind-tables
> diff --git a/elf/tst-unwind-main.c b/elf/tst-unwind-main.c
> new file mode 100644
> index 0000000000..d1236032d7
> --- /dev/null
> +++ b/elf/tst-unwind-main.c
> @@ -0,0 +1,38 @@
> +/* Test unwinding through main.
> +   Copyright (C) 2018 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +#include <unwind.h>
> +#include <unistd.h>
> +#include <support/test-driver.h>
> +
> +static _Unwind_Reason_Code
> +callback (struct _Unwind_Context *ctx, void *arg)
> +{
> +  return _URC_NO_REASON;
> +}
> +
> +int
> +main (void)
> +{
> +  /* Arrange for this test to be killed if _Unwind_Backtrace runs into an
> +     endless loop.  We cannot use the test driver because the complete
> +     call chain needs to be compiled with -funwind-tables so that
> +     _Unwind_Backtrace is able to reach _start.  */
> +  alarm (DEFAULT_TIMEOUT);
> +  _Unwind_Backtrace (callback, 0);
> +}
> diff --git a/sysdeps/riscv/start.S b/sysdeps/riscv/start.S
> index 4635ddb5eb..ba3e2cf16c 100644
> --- a/sysdeps/riscv/start.S
> +++ b/sysdeps/riscv/start.S
> @@ -43,6 +43,10 @@
>     __libc_start_main wants this in a5.  */
>
>  ENTRY (ENTRY_POINT)
> +	/* Terminate call stack by noting ra is undefined.  Use a dummy
> +	   .cfi_label to force starting the FDE.  */
> +	.cfi_label .Ldummy
> +	cfi_undefined (ra)
>  	call  .Lload_gp
>  	mv    a5, a0  /* rtld_fini.  */
>  	/* main may be in a shared library.  */
> @@ -54,7 +58,8 @@ ENTRY (ENTRY_POINT)
>  	lla   a4, __libc_csu_fini
>  	mv    a6, sp  /* stack_end.  */
>
> -	tail  __libc_start_main@plt
> +	call  __libc_start_main@plt
> +	unimp
>  END (ENTRY_POINT)
>
>  /* Dynamic links need the global pointer to be initialized prior to calling

GCC will emit an 'ebreak' when it sees unreachable code on RISC-V, for example

    $ cat test.c
    int main()
    {
      int *ptr = 0;
      return *ptr;
    }
    $ riscv64-sifive-elf-gcc test.c -O3 -S -o-
    	.file	"test.c"
    	.option nopic
    	.text
    	.section	.text.startup,"ax",@progbits
    	.align	1
    	.globl	main
    	.type	main, @function
    main:
    	lw	a5,0(zero)
    	ebreak
    	.size	main, .-main
    	.ident	"GCC: (GNU) 8.1.0"
    
Which results in a SIGBREAK instead of a SIGILL.  The decision here was a bit 
arbitrary, but I think we should keep it consistent between the hand-written 
assembly and the compiler-generated assembly.

Aside from that, this is great.  Thanks!