Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210714064129.1321277-1-victor.liu@nxp.com> <YO6qfQMcvr9szZTJ@smile.fi.intel.com>
 <1f1b38f9c42093bea5e6a8ccb458bdf799069157.camel@nxp.com> <YO6+nkzDlNM/KJio@smile.fi.intel.com>
 <YO7AcaHyB8js9FJg@smile.fi.intel.com> <0ae047e216298e6239e39ec40a2d545ac07a7406.camel@nxp.com>
 <CAHp75VfHuBJAYqms0tx_y3qjFfJJ_ynCOkbeYchBY0CvDJBp=Q@mail.gmail.com> <CAHp75VfFx4eDF484QKXgB=rhu3AptnDvX+2C+qd+U_3atfjnjQ@mail.gmail.com>
In-Reply-To: <CAHp75VfFx4eDF484QKXgB=rhu3AptnDvX+2C+qd+U_3atfjnjQ@mail.gmail.com>
From:   Andy Shevchenko <andy.shevchenko@gmail.com>
Date:   Thu, 15 Jul 2021 11:19:13 +0300
Message-ID: <CAHp75VchZmbd4BquLGnCAfvNVE-8dRoUULnWUFTJjNAiLnHHBA@mail.gmail.com>
Subject: Re: [RFC PATCH] clk: fractional-divider: Correct max_{m,n} handed
 over to rational_best_approximation()
To:     Liu Ying <victor.liu@nxp.com>,
        Mauro Carvalho Chehab <mchehab@kernel.org>
Cc:     Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        linux-clk <linux-clk@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Heikki Krogerus <heikki.krogerus@linux.intel.com>,
        Michael Turquette <mturquette@baylibre.com>,
        Stephen Boyd <sboyd@kernel.org>,
        Dong Aisheng <aisheng.dong@nxp.com>,
        NXP Linux Team <linux-imx@nxp.com>,
        Jacky Bai <ping.bai@nxp.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

+Cc: Mauro (below Q about sphinx and docs)

On Thu, Jul 15, 2021 at 11:14 AM Andy Shevchenko
<andy.shevchenko@gmail.com> wrote:
> On Thu, Jul 15, 2021 at 11:02 AM Andy Shevchenko
> <andy.shevchenko@gmail.com> wrote:

...

> Now I realize one more item which is missed in the big picture.
> When we have overflowed the denominator (n) and shifted the values, we
> are expecting that the caller will check the rate and use another
> 2^scale (see (6) below) prescaler if needed to drop frequency to the
> needed values. The first few users of this were the drivers of the IPs
> which have an additional prescaler themselves. I dunno if there is any
> flag in CCF to show that the set frequency is 2^scale higher than
> asked.
>
> It means if
>
>         r_o / r_u >> N_lim (5)
>
> we will get m and n *saturated* without this snipped, while with it
> they will be much much better with a nuance that resulting frequency
> is shifted left by
>
>         scale = ln2(r_o/r_u) - nlim (6)
>
> scale bits.

I think I have at some point to introduce a documentation and explain
all this from the thread there.

Mauro, is sphinx capable of producing TeX formulas?

-- 
With Best Regards,
Andy Shevchenko