[v2,0/4] kbuild: Per arch/platform dtc warning levels

Message ID 20231122-dtc-warnings-v2-0-bd4087325392@kernel.org
Headers
Series kbuild: Per arch/platform dtc warning levels |

Message

Rob Herring Nov. 22, 2023, 10:12 p.m. UTC
  This series adds support to set the dtc extra warning level on a per 
arch or per platform (directory really) basis.

The first version of this was just a simple per directory override for 
Samsung platforms, but Conor asked to be able to do this for all of 
riscv.

For merging, either I can take the whole thing or the riscv and samsung 
patches can go via their normal trees. The added variable will have no 
effect until merged with patch 2.

v1:
 - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/

Signed-off-by: Rob Herring <robh@kernel.org>
---
Rob Herring (4):
      kbuild: Move dtc graph_child_address warning to W=2
      kbuild: Allow arch/platform override of dtc warning level
      riscv: dts: Always enable extra W=1 warnings
      arm/arm64: dts: samsung: Always enable extra W=1 warnings

 arch/arm/boot/dts/samsung/Makefile  |  3 +++
 arch/arm64/boot/dts/exynos/Makefile |  3 +++
 arch/riscv/boot/dts/Makefile        |  3 +++
 scripts/Makefile.lib                | 16 +++++++++-------
 4 files changed, 18 insertions(+), 7 deletions(-)
---
base-commit: b85ea95d086471afb4ad062012a4d73cd328fa86
change-id: 20231122-dtc-warnings-968ff83a86ed

Best regards,
  

Comments

Masahiro Yamada Nov. 23, 2023, 7:38 a.m. UTC | #1
On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
>
> This series adds support to set the dtc extra warning level on a per
> arch or per platform (directory really) basis.
>
> The first version of this was just a simple per directory override for
> Samsung platforms, but Conor asked to be able to do this for all of
> riscv.
>
> For merging, either I can take the whole thing or the riscv and samsung
> patches can go via their normal trees. The added variable will have no
> effect until merged with patch 2.
>
> v1:
>  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
>
> Signed-off-by: Rob Herring <robh@kernel.org>
> ---


There were some attempts in the past to enable W=1 in particular subsystems,
so here is a similar comment.

Adding a new warning flag to W=1 is always safe without doing any compile test.

With this series, it would not be true any more because a new warning in W=1
would potentially break riscv/samsung platforms.

Linus requires a clean build (i.e. zero warning) when W= option is not given.






> Rob Herring (4):
>       kbuild: Move dtc graph_child_address warning to W=2
>       kbuild: Allow arch/platform override of dtc warning level
>       riscv: dts: Always enable extra W=1 warnings
>       arm/arm64: dts: samsung: Always enable extra W=1 warnings
>
>  arch/arm/boot/dts/samsung/Makefile  |  3 +++
>  arch/arm64/boot/dts/exynos/Makefile |  3 +++
>  arch/riscv/boot/dts/Makefile        |  3 +++
>  scripts/Makefile.lib                | 16 +++++++++-------
>  4 files changed, 18 insertions(+), 7 deletions(-)
> ---
> base-commit: b85ea95d086471afb4ad062012a4d73cd328fa86
> change-id: 20231122-dtc-warnings-968ff83a86ed
>
> Best regards,
> --
> Rob Herring <robh@kernel.org>
>
  
Conor Dooley Nov. 23, 2023, 9:43 a.m. UTC | #2
Yo,

On Thu, Nov 23, 2023 at 04:38:37PM +0900, Masahiro Yamada wrote:
> On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
> >
> > This series adds support to set the dtc extra warning level on a per
> > arch or per platform (directory really) basis.
> >
> > The first version of this was just a simple per directory override for
> > Samsung platforms, but Conor asked to be able to do this for all of
> > riscv.
> >
> > For merging, either I can take the whole thing or the riscv and samsung
> > patches can go via their normal trees. The added variable will have no
> > effect until merged with patch 2.
> >
> > v1:
> >  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
> >
> > Signed-off-by: Rob Herring <robh@kernel.org>
> > ---
> 
> 
> There were some attempts in the past to enable W=1 in particular subsystems,
> so here is a similar comment.
> 
> Adding a new warning flag to W=1 is always safe without doing any compile test.
> 
> With this series, it would not be true any more because a new warning in W=1
> would potentially break riscv/samsung platforms.

We carry a copy of the dtc in scripts/dtc, so I would expect that before
an upgrade is done that would introduce new warnings we can fix them in
the relevant platforms. Perhaps I am misunderstanding something there,
but I'm sure Rob can clear it up if I made a mistake.

> Linus requires a clean build (i.e. zero warning) when W= option is not given.

For RISC-V at least, there are currently no W=1 warnings while building
the dtbs (because I put effort into fixing them all) and I would like to
keep it that way, so that requirement is not a concern.

Cheers,
Conor.
  
Conor Dooley Nov. 23, 2023, 9:47 a.m. UTC | #3
On Wed, Nov 22, 2023 at 03:12:31PM -0700, Rob Herring wrote:
> This series adds support to set the dtc extra warning level on a per 
> arch or per platform (directory really) basis.
> 
> The first version of this was just a simple per directory override for 
> Samsung platforms, but Conor asked to be able to do this for all of 
> riscv.
> 
> For merging, either I can take the whole thing or the riscv and samsung 
> patches can go via their normal trees. The added variable will have no 
> effect until merged with patch 2.

IMO, you should take the lot.

Acked-by: Conor Dooley <conor.dooley@microchip.com>

Cheers,
Conor.
  
Masahiro Yamada Nov. 23, 2023, noon UTC | #4
On Thu, Nov 23, 2023 at 6:44 PM Conor Dooley <conor.dooley@microchip.com> wrote:
>
> Yo,
>
> On Thu, Nov 23, 2023 at 04:38:37PM +0900, Masahiro Yamada wrote:
> > On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
> > >
> > > This series adds support to set the dtc extra warning level on a per
> > > arch or per platform (directory really) basis.
> > >
> > > The first version of this was just a simple per directory override for
> > > Samsung platforms, but Conor asked to be able to do this for all of
> > > riscv.
> > >
> > > For merging, either I can take the whole thing or the riscv and samsung
> > > patches can go via their normal trees. The added variable will have no
> > > effect until merged with patch 2.
> > >
> > > v1:
> > >  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
> > >
> > > Signed-off-by: Rob Herring <robh@kernel.org>
> > > ---
> >
> >
> > There were some attempts in the past to enable W=1 in particular subsystems,
> > so here is a similar comment.
> >
> > Adding a new warning flag to W=1 is always safe without doing any compile test.
> >
> > With this series, it would not be true any more because a new warning in W=1
> > would potentially break riscv/samsung platforms.
>
> We carry a copy of the dtc in scripts/dtc, so I would expect that before
> an upgrade is done that would introduce new warnings we can fix them in
> the relevant platforms.


It might be easy to fix up only riscv and samsung platforms.

If more and more platforms add KBUILD_EXTRA_WARN=1 in their Makefiles,
it will become harder to sort them out before adding a new warning flag.


The sync of scripts/dtc will be delayed until then, or
you will end up with demoting them (i.e. delete KBUILD_EXTRA_WARN=1).



> Perhaps I am misunderstanding something there,
> but I'm sure Rob can clear it up if I made a mistake.
>
> > Linus requires a clean build (i.e. zero warning) when W= option is not given.
>
> For RISC-V at least, there are currently no W=1 warnings while building
> the dtbs (because I put effort into fixing them all) and I would like to
> keep it that way, so that requirement is not a concern.
>
> Cheers,
> Conor.
>
  
Rob Herring Nov. 27, 2023, 2:03 p.m. UTC | #5
On Thu, Nov 23, 2023 at 1:39 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
>
> On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
> >
> > This series adds support to set the dtc extra warning level on a per
> > arch or per platform (directory really) basis.
> >
> > The first version of this was just a simple per directory override for
> > Samsung platforms, but Conor asked to be able to do this for all of
> > riscv.
> >
> > For merging, either I can take the whole thing or the riscv and samsung
> > patches can go via their normal trees. The added variable will have no
> > effect until merged with patch 2.
> >
> > v1:
> >  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
> >
> > Signed-off-by: Rob Herring <robh@kernel.org>
> > ---
>
>
> There were some attempts in the past to enable W=1 in particular subsystems,
> so here is a similar comment.
>
> Adding a new warning flag to W=1 is always safe without doing any compile test.
>
> With this series, it would not be true any more because a new warning in W=1
> would potentially break riscv/samsung platforms.

The difference here is the people potentially adding warnings are also
the ones ensuring no warnings.

> Linus requires a clean build (i.e. zero warning) when W= option is not given.

Linus doesn't build any of this AFAICT. We are not always warning free
for W=0 with dtbs.

Rob
  
Masahiro Yamada Nov. 28, 2023, 12:02 p.m. UTC | #6
On Mon, Nov 27, 2023 at 11:03 PM Rob Herring <robh@kernel.org> wrote:
>
> On Thu, Nov 23, 2023 at 1:39 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> >
> > On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
> > >
> > > This series adds support to set the dtc extra warning level on a per
> > > arch or per platform (directory really) basis.
> > >
> > > The first version of this was just a simple per directory override for
> > > Samsung platforms, but Conor asked to be able to do this for all of
> > > riscv.
> > >
> > > For merging, either I can take the whole thing or the riscv and samsung
> > > patches can go via their normal trees. The added variable will have no
> > > effect until merged with patch 2.
> > >
> > > v1:
> > >  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
> > >
> > > Signed-off-by: Rob Herring <robh@kernel.org>
> > > ---
> >
> >
> > There were some attempts in the past to enable W=1 in particular subsystems,
> > so here is a similar comment.
> >
> > Adding a new warning flag to W=1 is always safe without doing any compile test.
> >
> > With this series, it would not be true any more because a new warning in W=1
> > would potentially break riscv/samsung platforms.
>
> The difference here is the people potentially adding warnings are also
> the ones ensuring no warnings.
>
> > Linus requires a clean build (i.e. zero warning) when W= option is not given.
>
> Linus doesn't build any of this AFAICT. We are not always warning free
> for W=0 with dtbs.



Does it mean, you can enable all warnings by default?






>
> Rob
>
  
Rob Herring Nov. 28, 2023, 10:25 p.m. UTC | #7
On Tue, Nov 28, 2023 at 6:03 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
>
> On Mon, Nov 27, 2023 at 11:03 PM Rob Herring <robh@kernel.org> wrote:
> >
> > On Thu, Nov 23, 2023 at 1:39 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> > >
> > > On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
> > > >
> > > > This series adds support to set the dtc extra warning level on a per
> > > > arch or per platform (directory really) basis.
> > > >
> > > > The first version of this was just a simple per directory override for
> > > > Samsung platforms, but Conor asked to be able to do this for all of
> > > > riscv.
> > > >
> > > > For merging, either I can take the whole thing or the riscv and samsung
> > > > patches can go via their normal trees. The added variable will have no
> > > > effect until merged with patch 2.
> > > >
> > > > v1:
> > > >  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
> > > >
> > > > Signed-off-by: Rob Herring <robh@kernel.org>
> > > > ---
> > >
> > >
> > > There were some attempts in the past to enable W=1 in particular subsystems,
> > > so here is a similar comment.
> > >
> > > Adding a new warning flag to W=1 is always safe without doing any compile test.
> > >
> > > With this series, it would not be true any more because a new warning in W=1
> > > would potentially break riscv/samsung platforms.
> >
> > The difference here is the people potentially adding warnings are also
> > the ones ensuring no warnings.
> >
> > > Linus requires a clean build (i.e. zero warning) when W= option is not given.
> >
> > Linus doesn't build any of this AFAICT. We are not always warning free
> > for W=0 with dtbs.
>
>
>
> Does it mean, you can enable all warnings by default?

No, Linus might not care, but others (me) do. The whole point of not
allowing warnings is the same. Get to zero warnings so any new
warnings stand out. We now have some subset of platforms which are
warning free and want warnings enabled by default to keep them that
way. How do you suggest we do that?

I understand your point on W=1 in general, but I think it just doesn't
apply in this case. In general,
someone may be testing a new compiler and there's some new warning to
enable, so they add it to W=1. They are working independently of any
subsystem (and Linus) and introducing new warnings would be a burden
to fix and a problem to leave. For DT, it is a bit different as adding
new warnings, updating dtc version, and selecting warnings to enable
are pretty much all done together. Plus, schema warnings have pretty
much superseded dtc warnings. If we do add new warnings which can't be
fixed up front, then we could still only enable the warning for W=1
from the command line. Something like this on top of this series:

diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 53a74e53e0ca..41307c6e1fee 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -341,6 +341,10 @@ quiet_cmd_gzip = GZIP    $@
 # ---------------------------------------------------------------------------
 DTC ?= $(objtree)/scripts/dtc/dtc

+ifeq ($(findstring 1,$(KBUILD_EXTRA_WARN)),)
+DTC_FLAGS += -Wno-some_new_warning_we_need_off_globally
+endif
+
 KBUILD_EXTRA_WARN_DTC += $(KBUILD_EXTRA_WARN)

 # Disable noisy checks by default
  
Masahiro Yamada Nov. 29, 2023, 5:49 p.m. UTC | #8
On Wed, Nov 29, 2023 at 7:25 AM Rob Herring <robh@kernel.org> wrote:
>
> On Tue, Nov 28, 2023 at 6:03 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> >
> > On Mon, Nov 27, 2023 at 11:03 PM Rob Herring <robh@kernel.org> wrote:
> > >
> > > On Thu, Nov 23, 2023 at 1:39 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> > > >
> > > > On Thu, Nov 23, 2023 at 7:12 AM Rob Herring <robh@kernel.org> wrote:
> > > > >
> > > > > This series adds support to set the dtc extra warning level on a per
> > > > > arch or per platform (directory really) basis.
> > > > >
> > > > > The first version of this was just a simple per directory override for
> > > > > Samsung platforms, but Conor asked to be able to do this for all of
> > > > > riscv.
> > > > >
> > > > > For merging, either I can take the whole thing or the riscv and samsung
> > > > > patches can go via their normal trees. The added variable will have no
> > > > > effect until merged with patch 2.
> > > > >
> > > > > v1:
> > > > >  - https://lore.kernel.org/all/20231116211739.3228239-1-robh@kernel.org/
> > > > >
> > > > > Signed-off-by: Rob Herring <robh@kernel.org>
> > > > > ---
> > > >
> > > >
> > > > There were some attempts in the past to enable W=1 in particular subsystems,
> > > > so here is a similar comment.
> > > >
> > > > Adding a new warning flag to W=1 is always safe without doing any compile test.
> > > >
> > > > With this series, it would not be true any more because a new warning in W=1
> > > > would potentially break riscv/samsung platforms.
> > >
> > > The difference here is the people potentially adding warnings are also
> > > the ones ensuring no warnings.
> > >
> > > > Linus requires a clean build (i.e. zero warning) when W= option is not given.
> > >
> > > Linus doesn't build any of this AFAICT. We are not always warning free
> > > for W=0 with dtbs.
> >
> >
> >
> > Does it mean, you can enable all warnings by default?
>
> No, Linus might not care, but others (me) do. The whole point of not
> allowing warnings is the same. Get to zero warnings so any new
> warnings stand out. We now have some subset of platforms which are
> warning free and want warnings enabled by default to keep them that
> way. How do you suggest we do that?





You may not like it, but an alternative solution could be,
hard-code extra warning flags.


In my compile-tests, Samsung platform is not W=1 clean yet.
I see -Wunit_address_vs_reg, -Wsimple_bus_reg,
-Wunique_unit_address_if_enabled warnings.

I do not see anything else, so you can add the following three
flags to keep it warning-free.






diff --git a/arch/arm/boot/dts/samsung/Makefile
b/arch/arm/boot/dts/samsung/Makefile
index 7becf36656b1..1e15784ec51f 100644
--- a/arch/arm/boot/dts/samsung/Makefile
+++ b/arch/arm/boot/dts/samsung/Makefile
@@ -1,4 +1,10 @@
 # SPDX-License-Identifier: GPL-2.0
+
+dtcflags := \
+       -Wavoid_unnecessary_addr_size \
+       -Walias_paths \
+       -Wgraph_child_address
+
 dtb-$(CONFIG_ARCH_EXYNOS3) += \
        exynos3250-artik5-eval.dtb \
        exynos3250-monk.dtb \
diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 1a965fe68e01..aa5a5fc39cec 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -362,6 +362,9 @@ DTC_FLAGS += -Wnode_name_chars_strict \
        -Wunique_unit_address
 endif

+# per-directory flags
+DTC_FLAGS += $(dtcflags)
+# per-file flags
 DTC_FLAGS += $(DTC_FLAGS_$(basetarget))

 # Set -@ if the target is a base DTB that overlay is applied onto







>
> I understand your point on W=1 in general, but I think it just doesn't
> apply in this case. In general,
> someone may be testing a new compiler and there's some new warning to
> enable, so they add it to W=1. They are working independently of any
> subsystem (and Linus) and introducing new warnings would be a burden
> to fix and a problem to leave. For DT, it is a bit different as adding
> new warnings, updating dtc version, and selecting warnings to enable
> are pretty much all done together.
> Plus, schema warnings have pretty
> much superseded dtc warnings. If we do add new warnings which can't be
> fixed up front, then we could still only enable the warning for W=1
> from the command line. Something like this on top of this series:
>
> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> index 53a74e53e0ca..41307c6e1fee 100644
> --- a/scripts/Makefile.lib
> +++ b/scripts/Makefile.lib
> @@ -341,6 +341,10 @@ quiet_cmd_gzip = GZIP    $@
>  # ---------------------------------------------------------------------------
>  DTC ?= $(objtree)/scripts/dtc/dtc
>
> +ifeq ($(findstring 1,$(KBUILD_EXTRA_WARN)),)
> +DTC_FLAGS += -Wno-some_new_warning_we_need_off_globally
> +endif
> +

Hmm. Tricky, but works.

KBUILD_EXTRA_WARN_DTC=1 is weaker than KBUILD_EXTRA_WARN=1




>  KBUILD_EXTRA_WARN_DTC += $(KBUILD_EXTRA_WARN)
>
>  # Disable noisy checks by default
>
--
Best Regards
Masahiro Yamada