Date:   Thu, 26 Jan 2023 10:30:24 -0800
From:   Ian Rogers <irogers@...gle.com>
To:     Josh Poimboeuf <jpoimboe@...nel.org>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        Nathan Chancellor <nathan@...nel.org>,
        Nick Desaulniers <ndesaulniers@...gle.com>,
        Tom Rix <trix@...hat.com>,
        Masahiro Yamada <masahiroy@...nel.org>,
        Nicolas Schier <nicolas@...sle.eu>,
        linux-kernel@...r.kernel.org, llvm@...ts.linux.dev,
        Stephane Eranian <eranian@...gle.com>,
        Andrii Nakryiko <andrii@...nel.org>,
        Jiri Olsa <jolsa@...nel.org>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        Namhyung Kim <namhyung@...nel.org>
Subject: Re: [PATCH v3 3/3] objtool: Alter how HOSTCC is forced

On Wed, Jan 25, 2023 at 5:49 PM Josh Poimboeuf <jpoimboe@...nel.org> wrote:
>
> On Thu, Jan 05, 2023 at 01:01:55AM -0800, Ian Rogers wrote:
> > HOSTCC is always wanted when building objtool. Setting CC to HOSTCC
> > happens after tools/scripts/Makefile.include is included, meaning
> > flags are set assuming say CC is gcc, but then it can be later set to
> > HOSTCC which may be clang. tools/scripts/Makefile.include is needed
> > for host set up and common macros in objtool's Makefile. Rather than
> > override CC to HOSTCC, just pass CC as HOSTCC to Makefile.build, the
> > libsubcmd builds and the linkage step. This means the Makefiles don't
> > see things like CC changing and tool flag determination, and similar,
> > work properly.
>
> I'm having trouble parsing this last sentence, can you rephrase or
> restructure it?

It was restating what was happening, deleted.

> > To avoid mixing CFLAGS from different compilers just
> > the objtool CFLAGS are determined with the exception of
> > EXTRA_WARNINGS.
>
> I'm not really sure what this one means either.

Moved to an inline comment.

> > HOSTCFLAGS is added to these so that command line
> > flags can add to the CFLAGS.
>
> Overall this patch description is a big wall of dense text which I found
> hard to decipher.  Please split it up into paragraphs and make it more
> legible and logical.

Yes, I've deleted most of the text now. I'd been adding to it as v1
went to v2 and so on.

> For example, un-jumble the ordering, with the background first, then the
> problem(s), then the fix(es).  (At least three paragraphs)
>
> If possible, the subject should describe the end result for the user,
> something like
>
>   objtool: Fix HOSTCFLAGS cmdline support
>
> ... unless I'm misunderstanding the point of the patch.

The patch is trying to fix the Makefile. Setting "CC=$(HOSTCC)" was
just wrong as apparent from looking at the behavior of
tools/scripts/Makefile.include. Some of that persists after this
change with WARNINGS, as now noted in a comment. A side effect of
fixing the Makefile is to make HOSTCFLAGS work, but I suspect with the
variable renames this may not be working. I'll leave that for yourself
and Nick. I told Nick I'd take a look at this as I saw the wrong use
of libsubcmd's headers and that was something I wanted to clean up
having done similar in tools/perf, along with the whole removal of
tools/lib/traceevent from Linux 6.2.

> HOSTCFLAGS from the cmdline *used* to work, a git bisect shows it
> stopped working with
>
>   96f14fe738b6 ("kbuild: Rename HOSTCFLAGS to KBUILD_HOSTCFLAGS")
>
> ... so please add a "Fixes" tag for that commit.

Left off, see later.

> > +MAKE = make -S
>
> Why?

Cruft, removed.

> >  LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
> >  ifneq ($(OUTPUT),)
> >    LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
> > @@ -37,12 +33,19 @@ INCLUDES := -I$(srctree)/tools/include \
> >           -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include \
> >           -I$(LIBSUBCMD_OUTPUT)/include
> >  WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
> > -CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
> > -LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
> > +OBJTOOL_CFLAGS := -Werror $(WARNINGS) -g $(INCLUDES) $(LIBELF_FLAGS) $(HOSTCFLAGS)
>
> I *think* I understand the reason for the switch from KBUILD_HOSTCFLAGS
> to HOSTCFLAGS -- because KBUILD_HOSTCFLAGS is defined in the top-level
> kernel Makefile which isn't included here so it's undefined? -- but
> regardless that should be called out more explicitly as a problem being
> fixed in the commit log.

I was matching tools/perf, I've switched back to KBUILD_HOSTCFLAGS in
v4. There's some higher logic at play with these variable names and
I'm not aware of it so I'll leave it to be fixed if necessary later.

> This issue really has me scratching my head, as I could have sworn
> objtool was being built with -O2.
>
> > +OBJTOOL_LDFLAGS := $(LIBELF_LIBS) $(LIBSUBCMD)
>
> Is there a reason to not include $(HOSTLDFLAGS) here?

Done as KBUILD_HOSTLDFLAGS as I don't see HOSTLDFLAGS set anywhere.

> >  # Allow old libelf to be used:
> >  elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E - | grep elf_getshdr)
> > -CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
> > +OBJTOOL_CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
> > +
> > +# Always want host compilation.
> > +HOST_OVERRIDES := CC="$(HOSTCC)" EXTRA_CFLAGS="$(OBJTOOL_CFLAGS)" \
> > +               LD="$(HOSTLD)" AR="$(HOSTAR)"
> > +BUILD_HOST_OVERRIDES := CC="$(HOSTCC)" CFLAGS="$(OBJTOOL_CFLAGS)" \
> > +                     LD="$(HOSTLD)" LDFLAGS="$(OBJTOOL_LDFLAGS)" \
> > +                     AR="$(HOSTAR)"
>
> Maybe it depends on your perspective, but I'm thinking that some of
> these aren't really overrides, but rather normal expected flags.  And
> the distinction between these two variables is muddy: it's not only
> Build vs non-Build, but also objtool vs libsubcmd.
>
> How about just
>
>   HOST_OVERRIDES := CC="$(HOSTCC) "LD="$(HOSTLD)" AR="$(HOSTAR)"
>
> And then specifying the {EXTRA_}CFLAGS and LDFLAGS below where needed.

Done.

> >  AWK = awk
> >  MKDIR = mkdir
> > @@ -65,10 +68,11 @@ include $(srctree)/tools/build/Makefile.include
> >
> >  $(OBJTOOL_IN): fixdep FORCE
> >       $(Q)$(CONFIG_SHELL) ./sync-check.sh
> > -     $(Q)$(MAKE) $(build)=objtool
> > +     $(Q)$(MAKE) $(build)=objtool $(BUILD_HOST_OVERRIDES)
> > +
> >
> >  $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
> > -     $(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
> > +     $(QUIET_LINK)$(HOSTCC) $(OBJTOOL_IN) $(KBUILD_HOSTLDFLAGS) $(OBJTOOL_LDFLAGS) -o $@
>
> Does KBUILD_HOSTLDFLAGS even work here?

Removed, albeit now to be part of OBJTOOL_LDFLAGS as your earlier
comment requested.

Thanks,
Ian

> --
> Josh

Powered by blists - more mailing lists