Re: mips build failures due to commit 8dd928915a73 (mips: fix up obsolete cpu function usage)

[Guenter Roeck]

On Tue, Apr 21, 2015 at 01:45:35PM +0930, Rusty Russell wrote:
> Aaro Koskinen <aaro.koskinen@xxxxxx> writes:
> > Hi,
> >
> > On Mon, Apr 20, 2015 at 12:40:28PM -0700, Guenter Roeck wrote:
> >> the upstream kernel fails to build mips:nlm_xlp_defconfig,
> >> mips:nlm_xlp_defconfig, mips:cavium_octeon_defconfig, and possibly
> >> other targets, with errors such as
> >> 
> >> arch/mips/kernel/smp.c:211:2: error:
> >> 	passing argument 2 of 'cpumask_set_cpu' discards 'volatile' qualifier
> >> 	from pointer target type
> >> arch/mips/kernel/process.c:52:2: error:
> >> 	passing argument 2 of 'cpumask_test_cpu' discards 'volatile' qualifier
> >> 	from pointer target type
> >> arch/mips/cavium-octeon/smp.c:242:2: error:
> >> 	passing argument 2 of 'cpumask_clear_cpu' discards 'volatile' qualifier
> >> 	from pointer target type
> >> 
> >> The problem was introduced with commit 8dd928915a73 (" mips: fix up
> >> obsolete cpu function usage"). I would send a patch to fix it, but I
> >> am not sure if removing 'volatile' from the variable declaration(s)
> >> would be a good idea.
> >
> > I think removing volatile from cpu_callin_map declaration should be OK,
> > since test_cpu (only reader) uses test_bit which takes care of it:
> >
> > 	static inline int test_bit(int nr, const volatile unsigned long *addr)
> 
> No, that got replaced too, with cpumask_test_cpu AFAICT.
> 
> You can open-code it, like so:
> 
>         test_bit(0, cpumask_bits(cpu_callin_map));
> 
> But you probably want to put a barrier in that loop instead of relying
> on volatile.
> 
The following might do it. Note that I can not really test it since I don't have
a real mips system, and qemu gets rcu hangs if I enable more than one CPU (I see
that with older kernels as well, so it is not a new problem). Someone will have
to test the patch on a real multi-core system.

Guenter

---