Re: [PATCH 3/6] M68K: cpuinfo: Fix a warning for CONFIG_CPUMASK_OFFSTACK
From: Huacai Chen
Date: Wed Jul 13 2022 - 22:07:29 EST
On Tue, Jul 12, 2022 at 6:15 PM WANG Xuerui <kernel@xxxxxxxxxx> wrote:
> Hi Geert and Huacai,
> On 2022/7/12 17:13, Geert Uytterhoeven wrote:
> > Hi Huacai,
> > On Tue, Jul 12, 2022 at 11:08 AM Huacai Chen <chenhuacai@xxxxxxxxx> wrote:
> >> On Tue, Jul 12, 2022 at 5:01 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> >>> On Tue, Jul 12, 2022 at 10:53 AM Huacai Chen <chenhuacai@xxxxxxxxx> wrote:
> >>>> On Tue, Jul 12, 2022 at 4:33 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> >>>>> On Tue, Jul 12, 2022 at 9:53 AM Huacai Chen <chenhuacai@xxxxxxxxxxx> wrote:
> >>>>>> When CONFIG_CPUMASK_OFFSTACK and CONFIG_DEBUG_PER_CPU_MAPS is selected,
> >>>>> DEBUG_PER_CPU_MAPS depends on SMP, which is not supported on m68k,
> >>>>> and thus cannot be enabled.
> >>>> This patch is derived from MIPS and LoongArch, I search all
> >>>> architectures and change those that look the same as MIPS and
> >>>> LoongArch.
> >>>> And the warning message below is also a copy-paste from LoongArch, sorry.
> >>>> Since M68K doesn't support SMP, then this patch seems to make no
> >>>> difference, but does it make sense to keep consistency across all
> >>>> architectures?
> >>> Yes, having consistency is good. But that should be mentioned in the
> >>> patch description, instead of a scary warning CCed to stable ;-)
> >>> BTW, you probably want to update the other copy of c_start() in
> >>> arch/m68k/kernel/setup_mm.c, too.
> >> For no-SMP architectures, it seems c_start() in
> >> arch/m68k/kernel/setup_mm.c is more reasonable (just use 1, neither
> >> NR_CPUS, nor nr_cpu_ids)?
> > The advantage of using nr_cpu_ids() is that this is one place less
> > to update when adding SMP support later...
> Hmm, so I've been watching m68k development lately (although not as
> closely as I'd like to, due to lack of vintage hardware at hand), given
> the current amazing momentum all the hobbyists/developers have been
> contributing to, SMP is well within reach...
> But judging from the intent of this patch series (fixing WARNs on
> certain configs), and that the triggering condition is currently
> impossible on m68k (and other non-SMP) platforms, I think cleanups for
> such arches could come as a separate patch series later. I think the
> m68k refactoring is reasonable after all, due to my observation above,
> but for the other non-SMP arches we may want to wait for the respective
> maintainers' opinions.
It seems that the best solution is only fix architectures with SMP
support and leave others (m68k, microblaze, um) as is. :)