Received-SPF: pass (google.com: domain of linux-kernel+bounces-87794-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Message-ID: <97787cc46d663b08b2e571841fb1bd6b.sboyd@kernel.org>
Content-Type: text/plain; charset="utf-8"
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
In-Reply-To: <SEZPR06MB69598A991B4CCF633E764CE7965F2@SEZPR06MB6959.apcprd06.prod.outlook.com>
References: <20240224-mux-v1-1-608cc704ef43@outlook.com> <0db752d31016823ccd3799925255a74a.sboyd@kernel.org> <SEZPR06MB69597E78EA08D5BDF66DBEBC965F2@SEZPR06MB6959.apcprd06.prod.outlook.com> <f8bcec654f618f8639a882bf70273618.sboyd@kernel.org> <SEZPR06MB69598A991B4CCF633E764CE7965F2@SEZPR06MB6959.apcprd06.prod.outlook.com>
Subject: Re: [PATCH RESEND] clk: set initial best mux parent to current parent when determining rate
From: Stephen Boyd <sboyd@kernel.org>
Cc: linux-clk@vger.kernel.org, linux-kernel@vger.kernel.org
To: Michael Turquette <mturquette@baylibre.com>, Yang Xiwen <forbidden405@outlook.com>, Yang Xiwen via B4 Relay <devnull+forbidden405.outlook.com@kernel.org>
Date: Thu, 29 Feb 2024 17:42:25 -0800
User-Agent: alot/0.10

Quoting Yang Xiwen (2024-02-28 18:33:11)
> On 2/29/2024 10:25 AM, Stephen Boyd wrote:
> >>>
> >>> Is the problem that we're not using abs_diff()?
> >>
> >>
> >> No, i think. It has nothing to do with the code here. It's because of
> >> the initial best_parent/best_parent_rate.
> >=20
> > Alright.

I will have to fix this as well in a different patch.

> >=20
> >>
> >>>
> >>> ----8<----
> >>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> >>> index a3bc7fb90d0f..91023345595f 100644
> >>> --- a/drivers/clk/clk.c
> >>> +++ b/drivers/clk/clk.c
> >>> @@ -542,7 +542,7 @@ static bool mux_is_better_rate(unsigned long rate=
, unsigned long now,
> >>>                          unsigned long best, unsigned long flags)
> >>>  {
> >>>       if (flags & CLK_MUX_ROUND_CLOSEST)
> >>> -             return abs(now - rate) < abs(best - rate);
> >>> +             return abs_diff(now, rate) < abs_diff(best, rate);
> >>
> >> Without this patch, the initial `best` rate would be always 0. This is
> >> wrong for most cases, 0Hz might (usually) be unavailable. We should use
> >> a valid rate(i.e. current rate) initially.
> >=20
> > Ok. But you set best to the parent rate. So why not use 'core->rate'
> > directly as 'best'?
>=20
>=20
> I can't remember exactly. I just add this piece of code and found it's
> working. Is this field already filled prior to setting rate? Anyway,
> your suggestion is very reasonable. Maybe dear clk maintainers can fix
> it as i'm not familiar with clk core code.

Yes the 'struct clk_rate_request' is pre-filled with many details,
including the rate of the clk and the current parent rate and parent hw
pointer. I'm pretty sure you're trying to fix this fixme from clk_test.c

static const struct clk_ops clk_dummy_single_parent_ops =3D {
	/*
	 * FIXME: Even though we should probably be able to use
	 * __clk_mux_determine_rate() here, if we use it and call
	 * clk_round_rate() or clk_set_rate() with a rate lower than
	 * what all the parents can provide, it will return -EINVAL.
	 *
	 * This is due to the fact that it has the undocumented
	 * behaviour to always pick up the closest rate higher than the
	 * requested rate. If we get something lower, it thus considers
	 * that it's not acceptable and will return an error.
	 *
	 * It's somewhat inconsistent and creates a weird threshold
	 * between rates above the parent rate which would be rounded to
	 * what the parent can provide, but rates below will simply
	 * return an error.
	 */