Re: PING^1 [PATCH] testsuite: 'b' instruction can't do long enough jumps

[Christophe Lyon <christophe.lyon@arm.com>]

On 9/28/22 15:39, Torbjorn SVENSSON wrote:
> Hi Christophe!
> 
> On 2022-09-28 13:55, Christophe Lyon wrote:
>> Hi!
>>
>>
>> On 9/28/22 11:17, Torbjorn SVENSSON via Gcc-patches wrote:
>>> Hi,
>>>
>>> Ping: 
>>> https://gcc.gnu.org/pipermail/gcc-patches/2022-September/601829.html
>>>
>>> Kind regards,
>>> Torbjörn
>>>
>>> On 2022-09-19 18:30, Torbjörn SVENSSON wrote:
>>>> After moving the testglue in commit 9d503515cee, the jump to exit and
>>>> abort is too far for the 'b' instruction on Cortex-M0. As most of the
>> I am not sure I understand why that commit changed the distance 
>> between 'exit' and the branch instruction?
> 
> The change was that the gcc_tg.o (the DejaGNU testglue.c object file) is 
> now put last on the command line. In the previous versions of GCC, it 
> was put before the ldflags flag etc, so now the code is placed in a way 
> that the distance might be too big.
> 
> This could also be related to that we in ST are using QEMU in system 
> mode and not user mode and as a result, our test environment is slightly 
> larger and might perhaps be placed in between the code for the test case 
> and the testglue.
> 

Thanks, that makes sense.

>>>> C code would generate a 'bl' instruction instead of a 'b'
>>>> instruction, lets do the same for the inline assembler.
>>>>
>>>> The error seen without this patch:
>>>>
>>>> /tmp/cccCRiCl.o: in function `main':
>>>> stack-protector-1.c:(.text+0x4e): relocation truncated to fit: 
>>>> R_ARM_THM_JUMP11 against symbol `__wrap_exit' defined in .text 
>>>> section in gcc_tg.o
>>>> stack-protector-1.c:(.text+0x50): relocation truncated to fit: 
>>>> R_ARM_THM_JUMP11 against symbol `__wrap_abort' defined in .text 
>>>> section in gcc_tg.o
>>>> collect2: error: ld returned 1 exit status
>>>>
>> Anyway the change seems sensible to me, I suppose it's not worth 
>> adding support in the linker to insert long branch stubs for these 
>> relocations.
> 
> If a simple 'bl' instead of 'b' is enough, I think that this trivial 
> change is the right one as the test case is supposed to test the stack 
> protection, not branching, right?
> 
Yeah, agreed, I just meant to say a linker patch in addition to this one 
is probably not worth the effort.

Christophe

> Kind regards,
> Torbjörn
> 
>>
>> Christophe
>>
>>>> gcc/testsuite/ChangeLog:
>>>>
>>>>          * gcc/testsuite/gcc.target/arm/stack-protector-1.c: Use 'bl'
>>>>     instead of 'b' instruction.
>>>>     * gcc/testsuite/gcc.target/arm/stack-protector-3.c: Likewise.
>>>>
>>>> Co-Authored-By: Yvan ROUX  <yvan.roux@foss.st.com>
>>>> Signed-off-by: Torbjörn SVENSSON  <torbjorn.svensson@foss.st.com>
>>>> ---
>>>>   gcc/testsuite/gcc.target/arm/stack-protector-1.c | 4 ++--
>>>>   gcc/testsuite/gcc.target/arm/stack-protector-3.c | 2 +-
>>>>   2 files changed, 3 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/gcc/testsuite/gcc.target/arm/stack-protector-1.c 
>>>> b/gcc/testsuite/gcc.target/arm/stack-protector-1.c
>>>> index 8d28b0a847c..3f0ffc9c3f3 100644
>>>> --- a/gcc/testsuite/gcc.target/arm/stack-protector-1.c
>>>> +++ b/gcc/testsuite/gcc.target/arm/stack-protector-1.c
>>>> @@ -56,8 +56,8 @@ asm (
>>>>   "    ldr    r1, [sp, #4]\n"
>>>>       CHECK (r1)
>>>>   "    mov    r0, #0\n"
>>>> -"    b    exit\n"
>>>> +"    bl    exit\n"
>>>>   "1:\n"
>>>> -"    b    abort\n"
>>>> +"    bl    abort\n"
>>>>   "    .size    main, .-main"
>>>>   );
>>>> diff --git a/gcc/testsuite/gcc.target/arm/stack-protector-3.c 
>>>> b/gcc/testsuite/gcc.target/arm/stack-protector-3.c
>>>> index b8f77fa2309..2f710529b8f 100644
>>>> --- a/gcc/testsuite/gcc.target/arm/stack-protector-3.c
>>>> +++ b/gcc/testsuite/gcc.target/arm/stack-protector-3.c
>>>> @@ -26,7 +26,7 @@ asm (
>>>>   "    .type    __stack_chk_fail, %function\n"
>>>>   "__stack_chk_fail:\n"
>>>>   "    movs    r0, #0\n"
>>>> -"    b    exit\n"
>>>> +"    bl    exit\n"
>>>>   "    .size    __stack_chk_fail, .-__stack_chk_fail"
>>>>   );