Re: linux-next: Tree for Sep 1 [sound/soc/mediatek/mt8195/snd-soc-mt8195-afe.ko]

[Chen-Yu Tsai]

On Fri, Sep 3, 2021 at 3:05 PM Trevor Wu <trevor.wu@xxxxxxxxxxxx> wrote:
>
> On Thu, 2021-09-02 at 17:15 +0800, Chen-Yu Tsai wrote:
> > On Thu, Sep 2, 2021 at 4:52 PM Trevor Wu <trevor.wu@xxxxxxxxxxxx>
> > wrote:
> > >
> > > On Thu, 2021-09-02 at 10:22 +0200, Geert Uytterhoeven wrote:
> > > > Hi Trevor,
> > > >
> > > > On Thu, Sep 2, 2021 at 4:37 AM Trevor Wu <trevor.wu@xxxxxxxxxxxx>
> > > > wrote:
> > > > > On Wed, 2021-09-01 at 13:55 -0700, Randy Dunlap wrote:
> > > > > > On 9/1/21 1:17 AM, Stephen Rothwell wrote:
> > > > > > > Please do not add any v5.16 related code to your linux-next
> > > > > > > included
> > > > > > > branches until after v5.15-rc1 has been released.
> > > > > > >
> > > > > > > Changes since 20210831:
> > > > > > >
> > > > > >
> > > > > >
> > > > > > on x86_64:
> > > > > >
> > > > > > ERROR: modpost: "clkdev_add" [sound/soc/mediatek/mt8195/snd-
> > > > > > soc-
> > > > > > mt8195-afe.ko] undefined!
> > > > > > ERROR: modpost: "clkdev_drop" [sound/soc/mediatek/mt8195/snd-
> > > > > > soc-
> > > > > > mt8195-afe.ko] undefined!
> > > > > > ERROR: modpost: "clk_unregister_gate"
> > > > > > [sound/soc/mediatek/mt8195/snd-
> > > > > > soc-mt8195-afe.ko] undefined!
> > > > > > ERROR: modpost: "clk_register_gate"
> > > > > > [sound/soc/mediatek/mt8195/snd-
> > > > > > soc-mt8195-afe.ko] undefined!
> > > > > >
> > > > > > Full randconfig file is attached.
> > > > > >
> > > > >
> > > > > Hi Randy,
> > > > >
> > > > > The problem is caused by the dependency declaration, because
> > > > > it's
> > > > > not a
> > > > > driver for x86_64.
> > > > > The dependency declaration has been added in the following
> > > > > patch.
> > > > >
> > > > >
> > >
> > >
> https://urldefense.com/v3/__https://patchwork.kernel.org/project/alsa-devel/patch/7e628e359bde04ceb9ddd74a45931059b4a4623c.1630415860.git.geert*renesas@xxxxxxxxx/__;Kw!!CTRNKA9wMg0ARbw!wMq130mAo-s45pP6ShQ1S8UIRuJLhwOnCbQNAQHIE2zvNhjAd67h1rlqkIDxJvC5_g$
> > > > >
> > > >
> > > > That is not sufficient, if COMPILE_TEST is enabled.
> > > >
> > > > Looks like it needs a dependency on COMMON_CLK, too.
> > > >
> > > > Gr{oetje,eeting}s,
> > > >
> > > >                         Geert
> > > >
> > >
> > > Hi Geert,
> > >
> > > Because it's a ARM64 driver, ARM64 will select COMMON_CLK.
> > > It seems that some dependency should be checked if COMPILE_TEST is
> > > enabled and the driver is compiled on non-ARM64 environment.
> > > We don't expect the driver can be used on non-ARM64 environment,may
> > > I
> > > remove COMPILE_TEST to solve the problem?
> > > If the driver only depends on ARCH_MEDIATEK, it must be compiled on
> > > ARM64.
> >
> > The whole point of COMPILE_TEST is that it gets compile-tested. It
> > doesn't
> > have to actually run.
> >
> > Since the driver is using parts of the common clk framework, it
> > should
> > declare an explicit dependency, instead of implicitly depending on
> > other symbols to enable it.
> >
> > ChenYu
>
> Hi ChenYu,
>
> Got it. I will send a patch for the problem.

I believe the background for this is that the kernel gets tested the most
on x86, even for build tests. We want code to get tested more, especially
for API changes for which the authors most likely do not have hardware
to test all combinations.

ChenYu