From mboxrd@z Thu Jan 1 00:00:00 1970 From: Viresh Kumar Subject: Re: [PATCH 1/2] cpufreq: exynos: use relaxed IO accesors Date: Wed, 22 Jun 2016 16:07:52 +0530 Message-ID: <20160622103752.GG5458@vireshk-i7> References: <1466587984-14105-1-git-send-email-ben.dooks@codethink.co.uk> <3668897.kYSEflAvDO@wuerfel> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-samsung-soc-owner@vger.kernel.org To: Krzysztof Kozlowski Cc: Arnd Bergmann , linux-arm-kernel@lists.infradead.org, linux-kernel@lists.codethink.co.uk, linux-samsung-soc@vger.kernel.org, linux-pm@vger.kernel.org, "Rafael J. Wysocki" , Ben Dooks List-Id: linux-pm@vger.kernel.org On 22-06-16, 12:36, Krzysztof Kozlowski wrote: > On Wed, Jun 22, 2016 at 11:53 AM, Arnd Bergmann wrote: > > On Wednesday, June 22, 2016 10:33:03 AM CEST Ben Dooks wrote: > >> The use of __raw IO accesors is not endian safe and should be used > >> sparingly. The relaxed variants should be as lightweight and also > >> are endian safe. > >> > >> Signed-off-by: Ben Dooks > >> > > > > Why not use the normal readl/writel() here instead of the relaxed version? > > > > Either one should work here, but in general I'd recommend using the > > non-relaxed version unless code is particularly performance sensitive. > > > > The main argument for that is to not let people get used to using > > _relaxed() all the time because it causes some very hard to debug > > problems in the cases where you actually need the barriers. > > I think that would be actually different patch, not endian related. > The concurrent operations here are excluded by mutexes so this looks > safe. > > Viresh, > I saw your ack. Do you prefer me to take the set through samsung-soc? Sure. -- viresh