From mboxrd@z Thu Jan  1 00:00:00 1970
From: sboyd@codeaurora.org (Stephen Boyd)
Date: Tue, 2 Jan 2018 10:16:08 -0800
Subject: [PATCH v6 4/5] clk: aspeed: Register gated clocks
In-Reply-To: <1514584997.2743.107.camel@kernel.crashing.org>
References: <20171128071908.12279-1-joel@jms.id.au>
 <20171128071908.12279-5-joel@jms.id.au>
 <20171221233927.GE7997@codeaurora.org>
 <1513910191.2743.77.camel@kernel.crashing.org>
 <1513910633.2743.79.camel@kernel.crashing.org>
 <20171227013227.GV7997@codeaurora.org>
 <1514584997.2743.107.camel@kernel.crashing.org>
Message-ID: <20180102181608.GG7997@codeaurora.org>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On 12/30, Benjamin Herrenschmidt wrote:
> On Tue, 2017-12-26 at 17:32 -0800, Stephen Boyd wrote:
> > > I noticed we do have a few i2c based clock drivers... how are they ever
> > > supposed to work ? i2c bus controllers are allowed to sleep and the i2c
> > > core takes mutexes...
> > 
> > We have clk_prepare()/clk_unprepare() for sleeping suckage. You
> > can use that, and i2c based clk drivers do that today.
> 
> "suckage" ? Hehe ... the suckage should rather be stuff that cannot
> sleep. Arbitrary latencies and jitter caused by too much code wanting
> to be "atomic" when unnecessary are a bad thing.

Heh. Of course.

> 
> In the case of clocks like the aspeed where we have to wait for a
> rather long stabilization delay, way too long to legitimately do a non-
> sleepable delay with a lock held, do we need to do everything in
> prepare() then ?
> 

Yes. If we have to wait a long time in the enable path it makes
sense to move it to the prepare path instead, if possible. That
way we avoid holding a spinlock for a long time.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project