Re: [PATCH] ARM: panic with a message when relocation type is unknown

[Andrey Smirnov <andrew.smirnov@gmail.com>]

On Mon, Jun 3, 2019 at 6:03 AM Ahmad Fatoum <a.fatoum@pengutronix.de> wrote:
>
> Hello Andrey,
>
> On 31/5/19 20:56, Andrey Smirnov wrote:
> > On Fri, May 31, 2019 at 2:02 AM Ahmad Fatoum <a.fatoum@pengutronix.de> wrote:
> >>
> >> Currently such failures result in a
> >>
> >>         >00000000 00000000
> >>
> >>         ### ERROR ### Please RESET the board ###
> >>
> >> With this patch this now becomes
> >>
> >>         >00000000 00000000
> >>         Unknown relocation type
> >>         ### ERROR ### Please RESET the board ###
> >>
> >> which improves user experience a little bit.
> >>
> >> Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
> >> ---
> >>  arch/arm/cpu/common.c | 4 ++--
> >>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/arch/arm/cpu/common.c b/arch/arm/cpu/common.c
> >> index 821cafbf26c2..3668c5977ca9 100644
> >> --- a/arch/arm/cpu/common.c
> >> +++ b/arch/arm/cpu/common.c
> >> @@ -78,7 +78,7 @@ void relocate_to_current_adr(void)
> >>                         putc_ll(' ');
> >>                         puthex_ll(rel->r_addend);
> >>                         putc_ll('\n');
> >> -                       panic("");
> >> +                       panic("Unknown relocation type");
> >>                 }
> >>
> >>                 dstart += sizeof(*rel);
> >> @@ -108,7 +108,7 @@ void relocate_to_current_adr(void)
> >>                         putc_ll(' ');
> >>                         puthex_ll(rel->r_offset);
> >>                         putc_ll('\n');
> >> -                       panic("");
> >> +                       panic("Unknown relocation type");
> >
> > Looking at this code makes me wonder if calls to panic() are even
> > appropriate here. I am not sure there's any guarantee that by the time
> > we get to this line relocation for "Unknown relocation type"(or
> > original "") string would be processed and panic() would get right
> > arguments not to mention all of the functions called as a part of
> > panic() and their potential dependencies. Another thing about painc()
> > is in PBL it completely ignores passed arguments, so no message will
> > be printed there.
> >
> > Maybe it'd make sense to go in a different direction and drop calls to
> > panic() altogether and replace them with:
> >
> > puts_ll(relocation_failed_message());
> > hang();
> >
> > where relocation_failed_message() on ARM can be something like:
> >
> > .section .text.relocation_failed_message
> > ENTRY(relocation_failed_message)
> > push {lr}
> > bl 1f
> > .byte 'U, 'n, 'k, 'n, 'o, 'w, 'n, '\ , 'r, 'e, 'l, 'o, 'c, 'a, 't, 'i,
> > 'o, 'n, '\ , 't, 'y, 'p, 'e, '\r, '\n, 0x00
> > 1:
> > mov r0, lr
> > /* In case we are in thumb */
> > bic r0, r0, #1
> > pop {pc}
> > ENDPROC(relocation_failed_message)
> >
> > to guarantee that it won't depend on relocation? Might be an overkill
> > though, so take this with a grain of salt.
>
> In my particular case, the string literal was already being accessed relative
> to the program counter.

I think that would really be arch dependent. If you were testing on
AArch64, which is much better about PC-relative addressing, then yeah,
that's what I'd expect. In the case of ARM however what we get is
(RDU2 in my case):

   7ad1a: 4805      ldr r0, [pc, #20] ; (7ad30 <relocate_to_current_adr+0x88>)
   7ad1c: f786 fa5c bl 11d8 <panic>
   7ad20: 000f9df0 .word 0x000f9df0
   7ad24: 000f9e20 .word 0x000f9e20
   7ad28: 000d4600 .word 0x000d4600
   7ad2c: 000f9df0 .word 0x000f9df0
   7ad30: 000a26be .word 0x000a26be

as well as a corresponding relocation:

arm-none-eabi-objdump -R barebox | grep 7ad30
0007ad30 R_ARM_RELATIVE    *ABS*

so, AFAICT, it's definitely going to have to be processed before it
can be used. More so, even if the right string address is passed, we
still would have to make sure that all of the functions that
implementation of panic() is using are not a subject to the same
problem. I think _that_ is a much bigger problem and that's why I
think we should move away from using panic() in that particular part
of the code base.

> Do you know if GCC can be coerced to always do that?

I don't think there's a way (to my knowledge, of course)  and I think
on ARM you'd be facing a fundamental limitation on how far PC-relative
addressing can reach. I tried placing the string right after the
function with __attribute__((section(".text.relocate_to_current_adr.data")))
but GCC didn't really pick it up and still used absolute address
constant.

Thanks,
Andrey Smirnov

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox