From: emilio@elopez.com.ar (=?ISO-8859-1?Q?Emilio_L=F3pez?=) Date: Mon, 16 Dec 2013 17:55:55 -0300 Subject: [PATCHv6 2/5] clk: socfpga: Add a clock type for the SD/MMC driver In-Reply-To: <20131215045116.23538.4@quantum> References: <1386880245-10192-1-git-send-email-dinguyen@altera.com> <1386880245-10192-3-git-send-email-dinguyen@altera.com> <201312142233.27143.arnd@arndb.de> <52AD1185.6090501@linaro.org> <20131215045116.23538.4@quantum> Message-ID: <52AF68DB.7090105@elopez.com.ar> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hi Mike et al, El 15/12/13 01:51, Mike Turquette escribi?: > clk_set_phase has been proposed before and now may be the time to add > it. There are two things that need to be addressed: > > 1) what are the values for the phase? This needs to work for others that > must set clk phase, so we need to consider all those requirements before > making a new function declaration in clk.h > > 2) is setting a clock's phase something done dynamically? Put another > way, does the same clock has it's phase set multiple times while the > system is running? For static configuration that only happens during > initialization we do not need a new API. The clock driver can handle it > privately. For dynamic operations though we likely need a new API. We on sunxi also need this for our (not yet merged) MMC driver; we currently have it implemented as an exported void clk_sunxi_mmc_phase_control(struct clk_hw *hw, u8 sample, u8 output) that takes the MMC clock (and only the MMC clock) and does the setup (it's basically configuring two values, "sample" and "output", into the clock register). I really don't know what does this do/why is it required/when is it used; I'm cc'ing Hans and David who can hopefully explain that part. I also saw a similar requirement from the gmac people (on cc too), who needed to set the phy type (or something like that) on one of the clock registers; so I'm thinking maybe a generic "tune something" approach that lets users configure some clock-dependent, arbitrary aspect of the clock is the way forward. Otherwise, we're going to end up bloating the clock framework with a lot of callback pointers that are going to be used in one or two clocks out of potentially hundreds on the whole system. Cheers, Emilio