Re: [PATCH] drm/i915: intel_backlight scale() math WA v2

["Eoff, Ullysses A" <ullysses.a.eoff@intel.com>]

On Mon, 2014-09-29 at 20:31 +0100, Damien Lespiau wrote:
> On Mon, Sep 29, 2014 at 05:50:57PM +0000, Eoff, Ullysses A wrote:
> > > -----Original Message-----
> > > From: Intel-gfx [mailto:intel-gfx-bounces@lists.freedesktop.org] On Behalf Of Jani Nikula
> > > Sent: Monday, September 29, 2014 6:07 AM
> > > To: Joe Konno; intel-gfx@lists.freedesktop.org
> > > Subject: Re: [Intel-gfx] [PATCH] drm/i915: intel_backlight scale() math WA v2
> > > 
> > > On Wed, 24 Sep 2014, Joe Konno <joe.konno@linux.intel.com> wrote:
> > > > From: Joe Konno <joe.konno@intel.com>
> > > >
> > > > Improper truncated integer division in the scale() function causes
> > > > actual_brightness != brightness. This (partial) work-around should be
> > > > sufficient for a majority of use-cases, but it is by no means a complete
> > > > solution.
> > > >
> > > > TODO: Determine how best to scale "user" values to "hw" values, and
> > > > vice-versa, when the ranges are of different sizes. That would be a
> > > > buggy scenario even with this work-around.
> > > >
> > > > The issue was introduced in the following (v3.17-rc1) commit:
> > > >
> > > >     6dda730 drm/i915: respect the VBT minimum backlight brightness
> > > >
> > > > v2: (thanks to Chris Wilson) clarify commit message, use rounded division
> > > > macro
> > > >
> > > > Signed-off-by: Joe Konno <joe.konno@intel.com>
> > > > ---
> > > >  drivers/gpu/drm/i915/intel_panel.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/gpu/drm/i915/intel_panel.c b/drivers/gpu/drm/i915/intel_panel.c
> > > > index f17ada3..dcdfbb3 100644
> > > > --- a/drivers/gpu/drm/i915/intel_panel.c
> > > > +++ b/drivers/gpu/drm/i915/intel_panel.c
> > > > @@ -421,7 +421,7 @@ static uint32_t scale(uint32_t source_val,
> > > >  	/* avoid overflows */
> > > >  	target_val = (uint64_t)(source_val - source_min) *
> > > >  		(target_max - target_min);
> > > > -	do_div(target_val, source_max - source_min);
> > > > +	target_val = DIV_ROUND_CLOSEST(target_val, source_max - source_min);
> > > 
> > > This fails to build with CONFIG_X86_32=y:
> > > 
> > > drivers/built-in.o: In function `scale':
> > > intel_panel.c:(.text+0x12ab38): undefined reference to `__udivdi3'
> > > make: *** [vmlinux] Error 1
> > > 
> > > 
> > 
> > Do you have a recommended workaround?  Should we just
> > use the v1 technique instead?
> 
> The problem is target_val is 64 bits and we're trying to do a 64 bits
> division with the 32bits instruction set. That is usually handled by
> __udivdi3 in libgcc (in userspace). We already have a
> DIV_ROUND_CLOSEST_ULL() that uses do_div() in intel_display.c.
> 

Ok, DIV_ROUND_CLOSEST_ULL() would be the one we want I take it.  Would
intel_drv.h be the appropriate header to move DIV_ROUND_CLOSEST_ULL()
into so that we can use it in intel_panel.c, too?

U. Artie