Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Arvind Sankar <nivedita@alum.mit.edu>
Date:   Thu, 6 Feb 2020 18:12:13 -0500
To:     Larry Finger <Larry.Finger@lwfinger.net>
Cc:     Ville =?utf-8?B?U3lyasOkbMOk?= <ville.syrjala@linux.intel.com>,
        Tom Anderson <thomasanderson@google.com>,
        Hans Verkuil <hansverk@cisco.com>,
        Manasi Navare <manasi.d.navare@intel.com>,
        LKML <linux-kernel@vger.kernel.org>
Subject: Re: Error building v5.5-git on PowerPC32 - bisected to commit
 7befe621ff81
Message-ID: <20200206231211.GA2976063@rani.riverdale.lan>
References: <0fb64c98-57c2-b988-051c-6ba0e460ad37@lwfinger.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <0fb64c98-57c2-b988-051c-6ba0e460ad37@lwfinger.net>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Feb 06, 2020 at 04:46:52PM -0600, Larry Finger wrote:
> When building post V5.5 on my PowerBook G4 Aluminum, the build failed with the 
> following error:
> 
> drivers/gpu/drm/drm_edid.c: In function ‘cea_mode_alternate_timings’:
> drivers/gpu/drm/drm_edid.c:3275:2: error: call to ‘__compiletime_assert_3282’ 
> declared with attribute error: BUILD_BUG_ON failed: cea_mode_for_vic(8)->vtotal 
> != 262 || cea_mode_for_vic(9)->vtotal != 262 || cea_mode_for_vic(12)->vtotal != 
> 262 || cea_mode_for_vic(13)->vtotal != 262 || cea_mode_for_vic(23)->vtotal != 
> 312 || cea_mode_for_vic(24)->vtotal != 312 || cea_mode_for_vic(27)->vtotal != 
> 312 || cea_mode_for_vic(28)->vtotal != 312
> 
> This error was bisected to commit 7befe621ff81 ("drm/edid: Abstract away 
> cea_edid_modes[]").
> 
> This problem is clearly a problem with the gcc compiler on the box, namely gcc 
> (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3. I had no success finding why the 
> attributes were wrong, and finally settled on the following hack:
> 
> diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
> index 99769d6c9f84..062bbe2b254a 100644
> --- a/drivers/gpu/drm/drm_edid.c
> +++ b/drivers/gpu/drm/drm_edid.c
> @@ -3272,6 +3272,7 @@ cea_mode_alternate_timings(u8 vic, struct drm_display_mode 
> *mode)
>           * get the other variants by simply increasing the
>           * vertical front porch length.
>           */
> +#ifndef CONFIG_PPC32
>          BUILD_BUG_ON(cea_mode_for_vic(8)->vtotal != 262 ||
>                       cea_mode_for_vic(9)->vtotal != 262 ||
>                       cea_mode_for_vic(12)->vtotal != 262 ||
> @@ -3280,6 +3281,7 @@ cea_mode_alternate_timings(u8 vic, struct drm_display_mode 
> *mode)
>                       cea_mode_for_vic(24)->vtotal != 312 ||
>                       cea_mode_for_vic(27)->vtotal != 312 ||
>                       cea_mode_for_vic(28)->vtotal != 312);
> +#endif
> 
>          if (((vic == 8 || vic == 9 ||
>                vic == 12 || vic == 13) && mode->vtotal < 263) ||
> 
> Disabling the build check allows me to build and test the post v5.5 versions.
> 
> Larry

It's not that the attributes are wrong. The problem is that BUILD_BUG_ON
requires a compile-time evaluatable condition. gcc-4.6 is apparently not
good enough at optimizing to reduce that expression to a constant,
though it was able to do it with the array accesses.