[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAP-5=fUx2+NaoHiNEqQzNPc7ODKaAM0e2fcf=O9XBC_r2HuPEw@mail.gmail.com>
Date: Fri, 2 May 2025 10:13:54 -0700
From: Ian Rogers <irogers@...gle.com>
To: Yury Norov <yury.norov@...il.com>
Cc: Rasmus Villemoes <linux@...musvillemoes.dk>, Arnd Bergmann <arnd@...db.de>,
Nathan Chancellor <nathan@...nel.org>, Nick Desaulniers <nick.desaulniers+lkml@...il.com>,
Bill Wendling <morbo@...gle.com>, Justin Stitt <justinstitt@...gle.com>,
Adrian Hunter <adrian.hunter@...el.com>, Thomas Gleixner <tglx@...utronix.de>,
Jakub Kicinski <kuba@...nel.org>, Jacob Keller <jacob.e.keller@...el.com>, linux-arch@...r.kernel.org,
linux-kernel@...r.kernel.org, llvm@...ts.linux.dev, Leo Yan <leo.yan@....com>
Subject: Re: [PATCH v2 2/5] bitmap: Silence a clang -Wshorten-64-to-32 warning
On Fri, May 2, 2025 at 9:55 AM Yury Norov <yury.norov@...il.com> wrote:
>
> On Fri, May 02, 2025 at 09:43:12AM -0700, Ian Rogers wrote:
> > On Fri, May 2, 2025 at 9:03 AM Yury Norov <yury.norov@...il.com> wrote:
> > >
> > > Hi Ian,
> > >
> > > On Wed, Apr 30, 2025 at 10:15:31AM -0700, Ian Rogers wrote:
> > > > The clang warning -Wshorten-64-to-32 can be useful to catch
> > > > inadvertent truncation. In some instances this truncation can lead to
> > > > changing the sign of a result, for example, truncation to return an
> > > > int to fit a sort routine. Silence the warning by making the implicit
> > > > truncation explicit. This isn't to say the code is currently incorrect
> > > > but without silencing the warning it is hard to spot the erroneous
> > > > cases.
> > > >
> > > > Signed-off-by: Ian Rogers <irogers@...gle.com>
> > > > ---
> > > > include/linux/bitmap.h | 2 +-
> > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/include/linux/bitmap.h b/include/linux/bitmap.h
> > > > index 595217b7a6e7..4395e0a618f4 100644
> > > > --- a/include/linux/bitmap.h
> > > > +++ b/include/linux/bitmap.h
> > > > @@ -442,7 +442,7 @@ static __always_inline
> > > > unsigned int bitmap_weight(const unsigned long *src, unsigned int nbits)
> > > > {
> > > > if (small_const_nbits(nbits))
> > > > - return hweight_long(*src & BITMAP_LAST_WORD_MASK(nbits));
> > > > + return (int)hweight_long(*src & BITMAP_LAST_WORD_MASK(nbits));
> > >
> > > This should return unsigned int, I guess?
> >
> > Hi Yury, I don't disagree. The issue there is that this could break
> > printf flags, etc. reliant on the return type. I've tried to keep the
> > patch minimal in this regard.
>
> Not sure I understand...
>
> I mean just
> return (unsigned int)hweight_long(...);
>
> because the bitmap_weight return type is unsigned int. Do I miss
> something?
Oh sorry, my misunderstanding.
> > > Also, most of the functions you touch here have their copies in tools.
> > > Can you please keep them synchronized?
> >
> > Yes, I do most of my work on the perf tool in the tools directory and
> > these patches come from adding -Wshorten-64-to-32 there due to a bug
> > found in ARM code that -Wshorten-64-to-32 would have caught:
> > https://lore.kernel.org/lkml/20250331172759.115604-1-leo.yan@arm.com/
> > The most recent patch series for tools is:
> > https://lore.kernel.org/linux-perf-users/20250430175036.184610-1-irogers@google.com/
> > However, I wanted to get the kernel versions of these headers agreed
> > before syncing them into the tools directory.
>
> Yes, I'm in CC for that series, but I didn't find the changes for
> bitmap_weight(), fls64() and other functions you touch in this series.
> Anyways, it would be logical to sync tools with the mother kernel in
> the same series.
Ok, I'll add it. Fwiw, I'm not particularly fond of syncing the files
as it's not clear how to do it and keep the attribution/changes clear.
I've patches to do things like make the tools/include more hermetic,
but they've died a death on LKML:
https://lore.kernel.org/lkml/20201015223119.1712121-1-irogers@google.com/
Tbh, I want to completely change how tools/include works as perf and
other things casually use -I into the tools/include and
tools/include/uapi directories meaning strange things like the types.h
in these things is usually the linux/types.h rather than
uapi/linux/types.h. Amongst other things, the licenses on these files
are not the same. I think we should be building the
tools/include/linux and associated tools/lib files as if they were a
library and installing their headers as is done for libbpf, libsubcmd,
etc. The -I would then reflect the install_headers output path. I'm
less clear on the value of doing this for uapi.
Thanks,
Ian
Powered by blists - more mailing lists