On Tue, May 05, 2026 at 12:46:16PM +0200, Quentin Schulz wrote: > Hi Simon, > > On 5/4/26 10:34 PM, Simon Glass wrote: > > Hi Quentin, > > > > On 2026-05-04T13:20:27, Quentin Schulz <[email protected]> wrote: > > > scripts/Makefile.extrawarn: sync with 6.1 > > > > > > Sync scripts/Makefile.extrawarn with Linux kernel 6.1's. > > > > > > Since it adds a new option for KBUILD_EXTRA_WARN with e, document it as > > > done in kernel commit c77d06e70d59 ("kbuild: support W=e to make build > > > abort in case of warning"). > > > > > > s/gcc/build/ in Makefile as done in kernel commit e27128db6283 ("kbuild: > > > rename KBUILD_ENABLE_EXTRA_GCC_CHECKS to KBUILD_EXTRA_WARN"), part of > > > v5.4. The rest of this commit was backported with commit bd3f9ee679b4 > > > ("kbuild: Bump the build system to 6.1"). > > > > > > Our scripts/Makefile.extrawarn before this commit is Linux kernel's from > > > v5.1 with the following diff: > > > > > > # diff --git a/proc/self/fd/11 b/proc/self/fd/13 > > > # --- a/proc/self/fd/11 > > > # +++ b/proc/self/fd/13 > > > # @@ -13,17 +13,8 @@ > > > [...] > > > > > > Makefile | 3 +- > > > scripts/Makefile.extrawarn | 154 > > > +++++++++++++++++++++++++++------------------ > > > 2 files changed, 95 insertions(+), 62 deletions(-) > > > > > diff --git a/scripts/Makefile.extrawarn b/scripts/Makefile.extrawarn > > > @@ -22,64 +17,101 @@ endif > > > ifdef CONFIG_CC_IS_CLANG > > > -KBUILD_CFLAGS += $(call cc-disable-warning, initializer-overrides) > > > -KBUILD_CFLAGS += $(call cc-disable-warning, unused-value) > > > -KBUILD_CFLAGS += $(call cc-disable-warning, format) > > > -KBUILD_CFLAGS += $(call cc-disable-warning, sign-compare) > > > -KBUILD_CFLAGS += $(call cc-disable-warning, format-zero-length) > > > -KBUILD_CFLAGS += $(call cc-disable-warning, uninitialized) > > > +KBUILD_CFLAGS += -Wno-initializer-overrides > > > +# Clang before clang-16 would warn on default argument promotions. > > > +ifneq ($(call clang-min-version, 160000),y) > > > > clang-min-version() is not defined anywhere that I can find. So I > > believe $(call clang-min-version, 160000) always expands to empty, the > > ifneq is always true and the inner ifeq always false. > > > > Thanks for catching this. It seems scripts/Makefile.compiler is also not > sync'ed with 6.1. We're missing kernel commit 88b61e3bff93 > ("Makefile.compiler: replace cc-ifversion with compiler-specific macros"). > I'll add that to the v2. > > > > diff --git a/scripts/Makefile.extrawarn b/scripts/Makefile.extrawarn > > > @@ -22,64 +17,101 @@ endif > > > -warning := $(warning-$(findstring 1, $(KBUILD_ENABLE_EXTRA_GCC_CHECKS))) > > > -warning += $(warning-$(findstring 2, $(KBUILD_ENABLE_EXTRA_GCC_CHECKS))) > > > -warning += $(warning-$(findstring 3, $(KBUILD_ENABLE_EXTRA_GCC_CHECKS))) > > > - > > > -ifeq ("$(strip $(warning))","") > > > - $(error W=$(KBUILD_ENABLE_EXTRA_GCC_CHECKS) is unknown) > > > -endif > > > > Dropping this silently accepts make W=q as a no-op. That matches v6.1 > > so I'm fine with it, but please note the behaviour change in the > > commit message so anyone bisecting a "my W=foo did nothing" report > > finds it. > > > > ACK. This was told in kernel commit 64a91907c896 ("kbuild: refactor > scripts/Makefile.extrawarn"): > > This commit slightly changes the behavior, I think all of which are OK. > > [1] Currently, cc-option calls are needlessly evaluated. For example, > warning-3 += $(call cc-option, -Wpacked-bitfield-compat) > needs evaluating only when W=3, but it is actually evaluated for > W=1, W=2 as well. With this commit, only relevant cc-option calls > will be evaluated. This is a slight optimization. > > [2] Currently, unsupported level like W=4 is checked by: > $(error W=$(KBUILD_ENABLE_EXTRA_GCC_CHECKS) is unknown) > This will no longer be checked, but I do not think it is a big > deal. > > [3] Currently, 4 Clang warnings (Winitializer-overrides, Wformat, > Wsign-compare, Wformat-zero-length) are shown by any of W=1, W=2, > and W=3. With this commit, they will be warned only by W=1. I > think this is a more correct behavior since each warning belongs > to only one group. > > > > diff --git a/scripts/Makefile.extrawarn b/scripts/Makefile.extrawarn > > > @@ -22,64 +17,101 @@ endif > > > -warning-1 += $(call cc-disable-warning, missing-field-initializers) > > > -warning-1 += $(call cc-disable-warning, sign-compare) > > > +KBUILD_CFLAGS += -Wmissing-prototypes > > > +KBUILD_CFLAGS += -Wold-style-definition > > > +KBUILD_CFLAGS += -Wmissing-include-dirs > > > > ... > > > +KBUILD_CFLAGS += -Wno-missing-field-initializers > > > +KBUILD_CFLAGS += -Wno-sign-compare > > > +KBUILD_CFLAGS += -Wno-type-limits > > > +KBUILD_CFLAGS += -Wno-shift-negative-value > > > > Several flags previously wrapped in cc-option / cc-disable-warning are > > now passed unconditionally (-Wmissing-prototypes, > > -Wold-style-definition, -Wmissing-include-dirs, -Wno-type-limits, > > -Wno-shift-negative-value). That's the v6.1 behaviour, fair enough. > > What's the oldest gcc we support? > > > > Pfew... That I don't know. We have an explicit check for GCC10 on ARM since > commit bb1c2b463265 ("update GCC version check after Kbuild bump"). We have > GCC4 min for PPC in arch/powerpc/config.mk with checkgcc4 target. > > mbedtls specifies they test against GCC 5.4, c.f. > https://github.com/Mbed-TLS/mbedtls#tool-versions. Some parts in > lib/mbedtls/external/mbedtls/library/aesce.c and > lib/mbedtls/external/mbedtls/library/sha256.c require GCC6. > lib/mbedtls/external/mbedtls/library/sha512.c requires GCC8. > > lwip's CI tests on ubuntu-latest GitHub runner, c.f. > https://github.com/lwip-tcpip/lwip/blob/master/.github/workflows/ci-linux.yml#L9-L17. > I don't know which version we've been using but it was merged for 2025.01 > and at that time ubuntu-latest GitHub worker was Ubuntu 22.04. I believe > GCC10 was the version by default, though 9 to 12 are available in the > Universe package feed. > > microMIPS requires GCC>=4.9, c.f. check in arch/mips/include/asm/compiler.h. > > include/linux/compiler-gcc.h requires GCC>=4.6. > > For the rest, I don't know. Maybe Tom does?
Absolute minimums are driven by other failures (for example, the bug we hit that pushed arm to gcc 10). GCC 10 is over 5 years old, so saying that's a global minimum is fine with me. And I've said before I'd be open to a global check. > The Linux kernel states in Documentation/process/changes.rst that the > minimum GCC is 5.1 though "The gcc version requirements may vary depending > on the type of CPU in your computer." (and I'm assuming "in your computer" > applies to the target device, not the host machine). I do wonder, idly, what platforms in the current kernel can build a functional kernel with 5.1 :) -- Tom
signature.asc
Description: PGP signature
