[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <NBKXPQ.SZZ17JHCOG5G@crapouillou.net>
Date: Sat, 13 Mar 2021 23:09:23 +0000
From: Paul Cercueil <paul@...pouillou.net>
To: Stephen Boyd <sboyd@...nel.org>
Cc: Michael Turquette <mturquette@...libre.com>,
Russell King <linux@...linux.org.uk>,
Ulf Hansson <ulf.hansson@...aro.org>, od@...c.me,
linux-clk@...r.kernel.org, linux-kernel@...r.kernel.org,
linux-mips@...r.kernel.org, linux-mmc@...r.kernel.org
Subject: Re: [PATCH 1/2] clk: Add clk_get_first_to_set_rate
Hi Stephen,
Le sam. 13 mars 2021 à 14:28, Stephen Boyd <sboyd@...nel.org> a écrit
:
> Quoting Paul Cercueil (2021-03-07 09:07:41)
>> The purpose of this function is to be used along with the notifier
>> mechanism.
>>
>> When a parent clock can see its rate externally changed at any
>> moment,
>> and a driver needs a specific clock rate to function, it can
>> register a
>> notifier on the parent clock, and call clk_set_rate() on the base
>> clock
>> to adjust its frequency according to the new parent clock.
>
> Can the driver use the rate locking mechanism to get a certain rate
> instead of registering for notifiers and trying to react to changes?
You mean with clk_rate_exclusive_get()? That sounds like a good idea,
but what would happen when a different driver calls the non-exclusive
clk_set_rate() on this clock (or the parent), would it return -EBUSY,
lock on a mutex? ...
Cheers,
-Paul
>
>>
>> This works fine, until the base clock has the CLK_SET_RATE_PARENT
>> flag
>> set. In that case, calling clk_set_rate() on the base clock will
>> call
>> clk_set_rate() on the parent clock, which will trigger the notifier
>> again, and we're in a loop.
>>
>> For that reason, we need to register the notifier on the parent
>> clock of
>> the first ancestor of the base clock that will effectively modify
>> its
>> rate when clk_set_rate() is called, which we can now obtain with
>> clk_get_first_to_set_rate().
>>
>> Signed-off-by: Paul Cercueil <paul@...pouillou.net>
Powered by blists - more mailing lists