Re: arch/m68k/include/asm/raw_io.h:91:13: warning: array subscript 0 is outside array bounds of 'volatile u16[0]' {aka 'volatile short unsigned int[]'}

[Michael Schmitz]

Hi Geert,

Am 21.09.23 um 19:08 schrieb Geert Uytterhoeven:
> Hi Günter, Rob,
>
> CC Michael
>
> On Wed, Sep 20, 2023 at 11:09 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>> On 9/20/23 09:20, Rob Herring wrote:
>> [....]
>>
>>>> Sure, but I still argue that this isn't worth it for drivers like this one.
>>>> Are you going to submit a fix ? Because otherwise I'll submit a patch
>>>> to drop COMPILE_TEST from MACHZ_WDT.
> I think dropping COMPILE_TEST from MACHZ_WDT is the right thing to do
> anyway.  Unlike most other drivers, this is not a driver that can be
> compiled in, and doesn't do anything if the hardware is not present.
> In fact it is a very dangerous driver: its probe function "reads" the
> ZF version register, but that involves doing an unconditional write,
> which might crash any non-X86 system.
>
> IMHO a driver must not be enabled for compile-testing if its presence
> can harm the system.
>
>>> I honestly don't know what the fix is. There's a compiler flag to
>>> allow 0 address, but that seems like a big hammer. From what I read on
>>> the fix for gcc-12, we shouldn't be getting this, but I haven't
>>> confirmed. I was hoping for comment from Geert as the issue doesn't
>>> appear to be the driver, but the arch code.
> Well, Atari ROM port ISA accesses are really weird, due to the really
> weird way the bus is wired to the address/data lines...

Putting it mildly ...

> The issue is that gcc considers accessing these addresses as "not
> done"...
That an new GCC problem?
>>> Furthermore, I just built the same HEAD and config as reported and
>>> don't see this error. I'm using kernel.org nolibc gcc 13.2.0 which
>>> should be the same version.
>> Exactly my point. So now we are stuck with a report like this on a
>> driver which probably has 0 users and we don't know how to fix it,
>> all to get the benefit of being able to compile it for an architecture
>> and platform which will never use it.
>>
>> I seem to recall similar errors with m68k and COMPILE_TEST last time
>> I tried to enable it on watchdog drivers, so I am not sure if this is
>> entirely new.
> Probably not.
>
> Michael: original build failure report in
> https://lore.kernel.org/r/202309192013.vI4DKHmw-lkp@xxxxxxxxx/

>From what I can see, DATA_W appears to be zero - maybe the driver needs
to set 'sane' values for IO ports if compile tested on non-ISA
architectures?

In the alternative, we might have to prevent setting ATARI_ROM_ISA if
COMPILE_TEST is set.

Cheers,

    Michael


>
> Gr{oetje,eeting}s,
>
>                         Geert
>