On Thu, May 31, 2018 at 12:00 AM, Andreas Färber <afaerber@xxxxxxx> wrote:
> Hi Luc,
>
> The typo in the subject made me curious...
>
> Am 30.05.2018 um 22:48 schrieb Luc Van Oostenryck:
>> By default, sparse assumes a 64bit machine when compiled on x86-64
>> and 32bit when compiled on anything else.
>>
>> This can of course create all sort of problems for the other archs, like
>> issuing false warnings ('shift too big (32) for type unsigned long'), or
>> worse, failing to emit legitimate warnings.
>>
>> Fix this by adding the -m32/-m64 flag, depending on CONFIG_64BIT,
>> to CHECKFLAGS in the main Makefile (and so for all archs).
>> Also, remove the now unneeded -m32/-m64 in arch specific Makefiles.
>>
>> Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@xxxxxxxxx>
>> ---
>> Makefile | 3 +++
>> arch/alpha/Makefile | 2 +-
>> arch/arm/Makefile | 2 +-
>> arch/arm64/Makefile | 2 +-
>> arch/ia64/Makefile | 2 +-
>> arch/mips/Makefile | 3 ---
>> arch/parisc/Makefile | 2 +-
>> arch/sparc/Makefile | 2 +-
>> arch/x86/Makefile | 2 +-
>> 9 files changed, 10 insertions(+), 10 deletions(-)
>
> What about the architectures not touched by your patch that previously
> had no -m32/-m64? (arc, c6x, h8300, hexagon, m68k, microblaze, nds32,
> nios2, openrisc, powerpc, riscv, s390, sh, unicore32, xtensa)
As explained in the patch, by default sparse uses -m64 if compiled on x86-64
and 32bit on everything else (well, more recent versions use -m64 if
compiled on any 64 bit machine). I think that most ppc devs use a ppc
machine and so ppc was most probably fine (at least ppc64) but I suspect
that most of these others archs either had never sparse used on them
or had a lot of wrong warnings. IOW, it was maybe OK but most probably
incorrect for them and now it is OK.
> You forgot to CC them on this patch.
I didn't thought/knew it was needed and the CC list is already
quite long but, if needed, no problem for me.
> Have you really checked that all their toolchains support the -m32/-m64
> flags you newly introduce for them? Apart from non-biarch architectures,
> I'm thinking of 31-bit s390 as a corner case where !64 != 32.
Hmm, there is no change to anything I call 'toolchain related', like
compiler and linker. The only change is sparse (or any other checker)
receiving now a correct and explicit -m32 or -m64.
For s390, as far as I know:
1) it has CONFIG_64BIT unconditionally definee (because the old 31bit
is no more supported, now everything is s390x only).
2) even if the *address space* was only 31 bit, I'm very sure
that sizeof(long) and sizeof(void*) was 4 on these machine
hence -m32 would have been correct.
Best regards,
-- Luc
