Received-SPF: pass (google.com: domain of linux-kernel+bounces-77531-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Message-ID: <c79909e4e55badc8f094d2ff8c4d34ca.sboyd@kernel.org>
Content-Type: text/plain; charset="utf-8"
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
In-Reply-To: <20240126201433.1830600-1-jan.dakinevich@salutedevices.com>
References: <20240126201433.1830600-1-jan.dakinevich@salutedevices.com>
Subject: Re: [PATCH] clk: allow to skip clk_core_req_round_rate_nolock()
From: Stephen Boyd <sboyd@kernel.org>
Cc: kernel@salutedevices.com, Jan Dakinevich <jan.dakinevich@salutedevices.com>
To: Jan Dakinevich <jan.dakinevich@salutedevices.com>, Michael Turquette <mturquette@baylibre.com>, linux-clk@vger.kernel.org, linux-kernel@vger.kernel.org
Date: Thu, 22 Feb 2024 15:20:49 -0800
User-Agent: alot/0.10

Quoting Jan Dakinevich (2024-01-26 12:14:33)
> Calling of clk_core_req_round_rate_nolock() can be time-consuming in a
> case of deep hierarchy with multiple dividers/parents. But if the clock
> already has exactly the same rate as desired, there is no need to
> determine how it could be rounded.

What exactly are you trying to avoid? Is this an optimization or a bug
fix? TL;DR: I'm unlikely to apply this patch.

I could see some driver implementing round_rate()/determine_rate() in a
way that rounds the rate passed in, so that even if the rate is what the
clk is running at _right now_, it still wants to change it to something
else, or at least call down into the driver to call the set_rate clk_op.
Applying this patch will break that. The contract is that
clk_set_rate(rate) =3D=3D clk_set_rate(clk_round_rate(rate)). It doesn't
look like anything needs to change.