Hi all,
A quick refresher: the clock framework APIs in include/linux/clk.h have allowed platforms to develop their own platform-specific implementations to manage clocks; this meant that everyone had their own definition of struct clk, duplicated much code and contributed negatively to the on-going quest for The One Image to Rule Them All.
The common clk framework is an attempt to define a generic struct clk which most platforms can use to build a clk tree and perform a well-defined set of operations against.
These five patches are the next iteration of the common clk framework. Since the V2 submission back in late September I ported the OMAP4 portion of OMAP's platform-specific clk framework and actively developed the generic code on a Panda board which revealed many bugs in V2.
The patches are based on Linus' v3.2-rc1 tag and can be pulled from: git://git.linaro.org/people/mturquette/linux.git http://git.linaro.org/gitweb?p=people/mturquette/linux.git%3Ba=shortlog%3Bh=...
A great deal of this work was first done by Jeremy Kerr, who in turn based his patches off of work by Ben Herrenschmidt (https://lkml.org/lkml/2011/5/20/81). Many others contributed to those patches and promptly had their work stolen by me. Thanks to all for their past contributions.
What to expect in this version:
.the most notable change is the removal of struct clk_hw. This extra layer of abstraction is only necessary if we want hide the definition of struct clk from platform code. Many developers expressed the need to know some details of the generic struct clk in the platform layer, and rightly so. Now struct clk is defined in include/linux/clk.h, protected by #ifdef CONFIG_GENERIC_CLK.
.flags have been introduced to struct clk, with several of them defined and used in the common code. These flags protect against changing clk rates or switching the clk parent while that clk is enabled; another flag is used to signal to clk_set_rate that it should ask the parent to change it's rate too.
.speaking of which, clk_set_rate has been overhauled and is now recursive. *collective groan*. clk_set_rate is still simple for the common case of simply setting a single clk's rate. But if your clk has the CLK_PARENT_SET_RATE flag and the .round_rate callback recommends changing the parent rate, then clk_set_rate will recurse upwards to the parent and try it all over again. In the event of a failure everything unwinds and all the clks go out for drinks.
.clk_register has been replaced by clk_init, which does NOT allocate memory for you. Platforms should allocate their own clk_hw_whatever structure which contains struct clk. clk_init is still necessary to initialize struct clk internals. clk_init also accepts struct device *dev as an argument, but does nothing with it. This is in anticipation of device tree support.
.Documentation! I'm sure somebody reads it.
.sysfs support. Visualize your clk tree at /sys/clk! Where would be a better place to put the clk tree besides the root of /sys/? When a consensus on this is reached I'll submit the proper changes to Documentation/ABI/testing/.
What's missing?
.per tree locking. I implemented this at the Linaro Connect conference but the implementation was unpopular, so it didn't make the cut. There needs to be better understanding of everyone's needs for this to work.
.rate change notifications. I simply didn't want to delay getting these patches to the list any longer, so the notifiers didn't make it in. I'll submit them to the list soon, or roll them into the V4 patchset. There are comments in the clk API definitions for where PRECHANGE, POSTCHANGE and ABORT propagation will go.
.basic mux clk, divider and dummy clk implementations. I think others have some code lying around to implement these, so I left them out.
.device tree support. I haven't looked much at the on-going discussions on the dt clk bindings. How compatible (or not) are the device tree clk bindings and the way these patches want to initialize clks?
.what is the overlap between common clk and clkdev? We're essentially tracking the clks in two places (common clk's tree and clkdevs's list), which feels a bit wasteful.
What else?
.OMAP4 support will be posted to LOML and LAKML in a separate patchset, since others might be interested in seeing a full port. It is a total hack, and is not ready for a formal submission.
Mike Turquette (5): clk: Kconfig: add entry for HAVE_CLK_PREPARE Documentation: common clk API clk: introduce the common clock framework clk: basic gateable and fixed-rate clks clk: export tree topology and clk data via sysfs
Documentation/clk.txt | 312 +++++++++++++++++++++++++++ drivers/clk/Kconfig | 24 ++ drivers/clk/Makefile | 5 +- drivers/clk/clk-basic.c | 208 ++++++++++++++++++ drivers/clk/clk-sysfs.c | 199 +++++++++++++++++ drivers/clk/clk.c | 541 +++++++++++++++++++++++++++++++++++++++++++++++ include/linux/clk.h | 199 +++++++++++++++++- 7 files changed, 1484 insertions(+), 4 deletions(-) create mode 100644 Documentation/clk.txt create mode 100644 drivers/clk/clk-basic.c create mode 100644 drivers/clk/clk-sysfs.c create mode 100644 drivers/clk/clk.c