On Thu, Feb 07, 2019 at 04:05:58PM +0100, Geert Uytterhoeven wrote:
> Hi Simon,
>
> On Thu, Feb 7, 2019 at 2:36 PM Simon Horman <horms+renesas@xxxxxxxxxxxx> wrote:
> > Support Z and Z2 clocks with parent frequencies greater than UINT32_MAX Hz
> > (~4.29GHz).
> >
> > The DIV_ROUND_CLOSEST_ULL() macro accepts a 64bit dividend and 32bit
> > divisor. This leads to truncation of the dividend, which is the Z or Z2
>
> truncation of the divisor.
>
> > parent clock frequency in HZ, on platforms where frequency of that clock is
> > greater than UINT32_MAX Hz.
> >
> > To resolve this problem the DIV64_U64_ROUND_CLOSEST() macro, which an
>
> which takes an
>
> > unsigned 32bit dividend and divisor, is introduced.
>
> 64-bit
>
> > An earlier version of this patch made use of the existing
> > DIV_ROUND_CLOSEST() macro, which accepts the prevailing type of the
> > dividend and divisor. However, this does not compile on 32bit systems, such
> > as i386 and mips, when called with the types used at this callsite, an
> > unsigned long long dividend and unsigned long divisor.
>
> Thanks for fixing this!
>
> > This work is in preparation for supporting the Z2 cloco on the R-Car Gen3
> > E3 (r8a77990) SoC which has a 4.8GHz parent clock.
> >
> > Signed-off-by: Simon Horman <horms+renesas@xxxxxxxxxxxx>
>
> With the above fixed:
> Reviewed-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
>
> > ---
> > v4: Add and use DIV64_U64_ROUND_CLOSEST
> >
> > v2: New patch
> > ---
> > drivers/clk/renesas/rcar-gen3-cpg.c | 4 ++--
> > include/linux/math64.h | 13 +++++++++++++
>
> While I have no issue taking this change through the clk-renesas tree if
> Andrew provides his Acked-by, I think the introduction of
> DIV64_U64_ROUND_CLOSEST() should be a separate patch.
>
> However, given commit 68600f623d69da42 ("mm: don't miss the last page
> because of round-off error") added both DIV64_U64_ROUND_UP() and its
> user under mm/, there's precedence for not splitting it off...
I also assumed a separate patch would be better,
but I found little evidence of that practice being
recently used in math64.h.
In any case, as I need to respin I'll break it out into
a separate patch.
