From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: From: Heiko =?ISO-8859-1?Q?St=FCbner?= To: Michael Turquette Cc: Jisheng Zhang , Stephen Boyd , linux-clk@vger.kernel.org, linux-arm-kernel@lists.infradead.org, Sebastian Hesselbarth , pi-cheng.chen@linaro.org Subject: Re: [Query]set clk rate must operate its coordinated clock Date: Mon, 06 Jun 2016 10:10:37 +0200 Message-ID: <2738343.ekabfXD0fs@diego> In-Reply-To: <20160311165152.4103.82090@quark.deferred.io> References: <20160309152420.20de187a@xhacker> <20160311165152.4103.82090@quark.deferred.io> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" List-ID: Hi Mike, Am Freitag, 11. M=E4rz 2016, 08:51:52 schrieb Michael Turquette: > Quoting Jisheng Zhang (2016-03-08 23:24:20) > > I have the following clk case which I dunno the elegant solution: > >=20 > >=20 > > cpuclk have two parents: cpupll and refclk. When set the cpuclk fre= q, we > > have to set its parent's freq, I.E cpupll freq. But before changing= the > > cpupll's freq, we should set its refclk as its parent firstly. > >=20 > > AFAIK, this is a common case, I have seen such requirement in rockc= hip, > > samsung clk driver. They solve this by notifier, but as pointed out= by > > Michael in > > http://lists.infradead.org/pipermail/linux-arm-kernel/2015-June/351= 565.ht > > ml > >=20 > > "This is also a hack and it points towards some missing infrastruct= ure in > > the clock framework." > >=20 > > I also don't like the notifier solution, I believe the elegant solu= tion > > could be using the coordinated clock infrastructure. So what's the = status > > of this infrastructure? I can test, and I can even add some code to= make > > it be ready to be merged if you guide me ;) >=20 > Thanks for the email. I hope to finish that feature before ELC. I hav= e > Cc'd Pi-Cheng Chen from Mediatek who is also interested in coordinate= d > clock rates. pulling this up again, did this materialize yet? :-) Because it looks like on Rockchip we're getting yet another clock that = needs=20 special handling (set_rate needs to go through the trusted-firmware, wh= ile=20 reading stays the same) which might profit from that as well. Heiko