[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <f16f97d2-6f2f-9c5d-644b-d1c3470de06f@linux.intel.com>
Date: Wed, 4 Sep 2024 16:37:41 +0300 (EEST)
From: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
To: Gergo Koteles <soyer@....hu>
cc: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
Hans de Goede <hdegoede@...hat.com>, platform-driver-x86@...r.kernel.org,
LKML <linux-kernel@...r.kernel.org>, Ike Panhc <ike.pan@...onical.com>,
Peter Zijlstra <peterz@...radead.org>,
Josh Poimboeuf <jpoimboe@...nel.org>,
Nathan Chancellor <nathan@...nel.org>, kernel test robot <lkp@...el.com>
Subject: Re: [PATCH v1 1/1] platform/x86: ideapad-laptop: Make the scope_guard()
clear of its scope
On Tue, 3 Sep 2024, Gergo Koteles wrote:
> On Tue, 2024-09-03 at 18:14 +0300, Andy Shevchenko wrote:
> > On Tue, Sep 03, 2024 at 05:00:51PM +0200, Gergo Koteles wrote:
> > > On Thu, 2024-08-29 at 19:50 +0300, Andy Shevchenko wrote:
> > > > First of all, it's a bit counterintuitive to have something like
> > > >
> > > > int err;
> > > > ...
> > > > scoped_guard(...)
> > > > err = foo(...);
> > > > if (err)
> > > > return err;
> > > >
> > > > Second, with a particular kernel configuration and compiler version in
> > > > one of such cases the objtool is not happy:
> > > >
> > > > ideapad-laptop.o: warning: objtool: .text.fan_mode_show: unexpected end of section
> > > >
> > > > I'm not an expert on all this, but the theory is that compiler and
> > > > linker in this case can't understand that 'result' variable will be
> > > > always initialized as long as no error has been returned. Assigning
> > > > 'result' to a dummy value helps with this. Note, that fixing the
> > > > scoped_guard() scope (as per above) does not make issue gone.
> > > >
> > > > That said, assign dummy value and make the scope_guard() clear of its scope.
> > > > For the sake of consistency do it in the entire file.
> > > >
> > >
> > > Interestingly, if I open a scope manually and use the plain guard, the
> > > warning disappears.
> >
> > Yes, that's what I also have, but I avoid that approach because in that case
> > the printing will be done inside the lock, widening the critical section for
> > no benefits.
> >
>
> This is intended to be an inner block scope within the function, it
> does not expand the critical section.
>
>
> > > ...
> > > unsigned long result;
> > > int err;
> > >
> > > {
> > > guard(mutex)(&priv->vpc_mutex);
> > > err = read_ec_data(priv->adev->handle, VPCCMD_R_FAN,
> > > &result);
> > > if (err)
> > > return err;
> > > }
> > > ...
> > >
> > > This looks a bit strange, but is probably easier for the compiler than
> > > the for loop of scoped_guard.
> > >
> > > But I don't know how well this style fits into the kernel.
It's ugly enough that I'd prefer the initialization of results variable as
done in Andy's patch.
--
i.
Powered by blists - more mailing lists