Re: [PATCH] checkpatch: Add exceptions for dsb keyword usage

From: Joe Perches
Date: Fri Jul 06 2018 - 01:53:10 EST


On Fri, 2018-07-06 at 06:45 +0100, Mark Rutland wrote:
> On Thu, Jul 05, 2018 at 02:14:28PM -0700, Joe Perches wrote:
> > On Thu, 2018-07-05 at 11:19 -0700, Prakruthi Deepak Heragu wrote:
> > > mb() API can relpace the dsb() API in the kernel code. So, dsb() usage
> > > is discouraged. However, there are exceptions when dsb is used in a
> > > variable or a function name. Exceptions are when 'dsb' is prefixed with
> > > class [-_>*\.] and/or suffixed with class [-_\.;].
>
> This is a really confusing way of describing the match behaviour, and doesn't
> explain why this is a big problem.
>
> In C it's either:
>
> dsb()
> dsb(scope) // e.g. dsb(ish)
>
> ... where scope is [a-z]*.
>
> ... which can be matched as something like 'dsb([a-z]*)' if necessary.
>
> > []
> > > diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> >
> > []
> > > @@ -5372,6 +5372,12 @@ sub process {
> > > "Avoid line continuations in quoted strings\n" . $herecurr);
> > > }
> > >
> > > +# dsb is too ARMish, and should usually be mb.
> > > + if ($line =~ /[^-_>*\.]\bdsb\b[^-_\.;]/) {
> > > + WARN("ARM_BARRIER",
> > > + "Use of dsb is discouranged: prefer mb.\n" .
> > > + $herecurr);
> > > + }
> >
> > This patch is whitespace damaged with a spelling error.
> >
> > Also, if this is reasonable test, and I don't know
> > that it is, it should be cc'd to the linux-arm list
> > linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> >
> > Also, I suggest 2 tests, one for .S files and
> > another for .[ch] files, and this be made specific
> > to arch/arm... files
> >
> > Something like:
> >
> > if ($realfile =~ @^arch/arm@ &&
> > ($realfile =~ /\.S$/ && $line =~ /\bdsb\b/) ||
> > ($realfile =~ /\.[ch]$/ && $line =~ /\bdsb\s*\(/)) {
> > WARN("ARM_DSB",
> > "Prefer mb over dsb as an ARM memory barrier\n" . $herecurr);
> > }
> >
> > ARM people, is this reasonable?
>
> I don't think this is a big deal today.
>
> For code under arch/{arm,arm64}, it's perfectly reasonable to use dsb.
>
> For code *ouside* of arch/{arm,arm64}, there are a number of cases where we
> want to use dsb(), e.g. when dealing with architectural drivers that require
> special barriers, or for common code shared across arm and arm64.
>
> It doesn't look like this is a big problem today, anyhow:
>
> [mark@salmiak:~/src/linux]% git grep -w 'dsb(.*)' -- ^arch
> drivers/irqchip/irq-armada-370-xp.c: dsb();
> drivers/irqchip/irq-gic-v3-its.c: dsb(ishst);
> drivers/irqchip/irq-gic-v3-its.c: dsb(ishst);
> drivers/irqchip/irq-gic-v3-its.c: dsb(sy);
> drivers/irqchip/irq-gic-v3-its.c: dsb(sy);
> drivers/irqchip/irq-gic-v3-its.c: dsb(sy);
> drivers/mtd/nand/raw/cmx270_nand.c: dsb();
> drivers/mtd/nand/raw/cmx270_nand.c: dsb();
> drivers/mtd/nand/raw/cmx270_nand.c: dsb();
> drivers/mtd/nand/raw/cmx270_nand.c: dsb();
> drivers/mtd/nand/raw/cmx270_nand.c: dsb();
> drivers/perf/arm_spe_pmu.c: dsb(nsh);
> drivers/perf/arm_spe_pmu.c: dsb(nsh);
> drivers/power/reset/arm-versatile-reboot.c: dsb();
> drivers/soc/rockchip/pm_domains.c: dsb(sy);
> drivers/soc/rockchip/pm_domains.c: dsb(sy);
> drivers/staging/mt7621-mmc/sd.c: //dsb(); /* --- by chhung */
> drivers/staging/mt7621-mmc/sd.c: //dsb(); /* --- by chhung */
> drivers/staging/vc04_services/interface/vchiq_arm/vchiq.h:#define dsb(a)
> drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c: dsb(sy); /* data barrier operation */
> drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.c: dsb(sy);
> drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h: do { debug_ptr[DEBUG_ ## d] = __LINE__; dsb(sy); } while (0)
> drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h: do { debug_ptr[DEBUG_ ## d] = (v); dsb(sy); } while (0)
> drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h: do { debug_ptr[DEBUG_ ## d]++; dsb(sy); } while (0)
> virt/kvm/arm/hyp/vgic-v3-sr.c: dsb(sy);
> virt/kvm/arm/hyp/vgic-v3-sr.c: dsb(sy);
> virt/kvm/arm/hyp/vgic-v3-sr.c: dsb(sy);

Thanks Mark.

So it seems this shouldn't be applied.