1========================
2The Common Clk Framework
3========================
4
5:Author: Mike Turquette <mturquette@ti.com>
6
7This document endeavours to explain the common clk framework details,
8and how to port a platform over to this framework.  It is not yet a
9detailed explanation of the clock api in include/linux/clk.h, but
10perhaps someday it will include that information.
11
12Introduction and interface split
13================================
14
15The common clk framework is an interface to control the clock nodes
16available on various devices today.  This may come in the form of clock
17gating, rate adjustment, muxing or other operations.  This framework is
18enabled with the CONFIG_COMMON_CLK option.
19
20The interface itself is divided into two halves, each shielded from the
21details of its counterpart.  First is the common definition of struct
22clk which unifies the framework-level accounting and infrastructure that
23has traditionally been duplicated across a variety of platforms.  Second
24is a common implementation of the clk.h api, defined in
25drivers/clk/clk.c.  Finally there is struct clk_ops, whose operations
26are invoked by the clk api implementation.
27
28The second half of the interface is comprised of the hardware-specific
29callbacks registered with struct clk_ops and the corresponding
30hardware-specific structures needed to model a particular clock.  For
31the remainder of this document any reference to a callback in struct
32clk_ops, such as .enable or .set_rate, implies the hardware-specific
33implementation of that code.  Likewise, references to struct clk_foo
34serve as a convenient shorthand for the implementation of the
35hardware-specific bits for the hypothetical "foo" hardware.
36
37Tying the two halves of this interface together is struct clk_hw, which
38is defined in struct clk_foo and pointed to within struct clk_core.  This
39allows for easy navigation between the two discrete halves of the common
40clock interface.
41
42Common data structures and api
43==============================
44
45Below is the common struct clk_core definition from
46drivers/clk/clk.c, modified for brevity::
47
48	struct clk_core {
49		const char		*name;
50		const struct clk_ops	*ops;
51		struct clk_hw		*hw;
52		struct module		*owner;
53		struct clk_core		*parent;
54		const char		**parent_names;
55		struct clk_core		**parents;
56		u8			num_parents;
57		u8			new_parent_index;
58		...
59	};
60
61The members above make up the core of the clk tree topology.  The clk
62api itself defines several driver-facing functions which operate on
63struct clk.  That api is documented in include/linux/clk.h.
64
65Platforms and devices utilizing the common struct clk_core use the struct
66clk_ops pointer in struct clk_core to perform the hardware-specific parts of
67the operations defined in clk-provider.h::
68
69	struct clk_ops {
70		int		(*prepare)(struct clk_hw *hw);
71		void		(*unprepare)(struct clk_hw *hw);
72		int		(*is_prepared)(struct clk_hw *hw);
73		void		(*unprepare_unused)(struct clk_hw *hw);
74		int		(*enable)(struct clk_hw *hw);
75		void		(*disable)(struct clk_hw *hw);
76		int		(*is_enabled)(struct clk_hw *hw);
77		void		(*disable_unused)(struct clk_hw *hw);
78		unsigned long	(*recalc_rate)(struct clk_hw *hw,
79						unsigned long parent_rate);
80		long		(*round_rate)(struct clk_hw *hw,
81						unsigned long rate,
82						unsigned long *parent_rate);
83		int		(*determine_rate)(struct clk_hw *hw,
84						  struct clk_rate_request *req);
85		int		(*set_parent)(struct clk_hw *hw, u8 index);
86		u8		(*get_parent)(struct clk_hw *hw);
87		int		(*set_rate)(struct clk_hw *hw,
88					    unsigned long rate,
89					    unsigned long parent_rate);
90		int		(*set_rate_and_parent)(struct clk_hw *hw,
91					    unsigned long rate,
92					    unsigned long parent_rate,
93					    u8 index);
94		unsigned long	(*recalc_accuracy)(struct clk_hw *hw,
95						unsigned long parent_accuracy);
96		int		(*get_phase)(struct clk_hw *hw);
97		int		(*set_phase)(struct clk_hw *hw, int degrees);
98		void		(*init)(struct clk_hw *hw);
99		void		(*debug_init)(struct clk_hw *hw,
100					      struct dentry *dentry);
101	};
102
103Hardware clk implementations
104============================
105
106The strength of the common struct clk_core comes from its .ops and .hw pointers
107which abstract the details of struct clk from the hardware-specific bits, and
108vice versa.  To illustrate consider the simple gateable clk implementation in
109drivers/clk/clk-gate.c::
110
111	struct clk_gate {
112		struct clk_hw	hw;
113		void __iomem    *reg;
114		u8              bit_idx;
115		...
116	};
117
118struct clk_gate contains struct clk_hw hw as well as hardware-specific
119knowledge about which register and bit controls this clk's gating.
120Nothing about clock topology or accounting, such as enable_count or
121notifier_count, is needed here.  That is all handled by the common
122framework code and struct clk_core.
123
124Let's walk through enabling this clk from driver code::
125
126	struct clk *clk;
127	clk = clk_get(NULL, "my_gateable_clk");
128
129	clk_prepare(clk);
130	clk_enable(clk);
131
132The call graph for clk_enable is very simple::
133
134	clk_enable(clk);
135		clk->ops->enable(clk->hw);
136		[resolves to...]
137			clk_gate_enable(hw);
138			[resolves struct clk gate with to_clk_gate(hw)]
139				clk_gate_set_bit(gate);
140
141And the definition of clk_gate_set_bit::
142
143	static void clk_gate_set_bit(struct clk_gate *gate)
144	{
145		u32 reg;
146
147		reg = __raw_readl(gate->reg);
148		reg |= BIT(gate->bit_idx);
149		writel(reg, gate->reg);
150	}
151
152Note that to_clk_gate is defined as::
153
154	#define to_clk_gate(_hw) container_of(_hw, struct clk_gate, hw)
155
156This pattern of abstraction is used for every clock hardware
157representation.
158
159Supporting your own clk hardware
160================================
161
162When implementing support for a new type of clock it is only necessary to
163include the following header::
164
165	#include <linux/clk-provider.h>
166
167To construct a clk hardware structure for your platform you must define
168the following::
169
170	struct clk_foo {
171		struct clk_hw hw;
172		... hardware specific data goes here ...
173	};
174
175To take advantage of your data you'll need to support valid operations
176for your clk::
177
178	struct clk_ops clk_foo_ops = {
179		.enable		= &clk_foo_enable,
180		.disable	= &clk_foo_disable,
181	};
182
183Implement the above functions using container_of::
184
185	#define to_clk_foo(_hw) container_of(_hw, struct clk_foo, hw)
186
187	int clk_foo_enable(struct clk_hw *hw)
188	{
189		struct clk_foo *foo;
190
191		foo = to_clk_foo(hw);
192
193		... perform magic on foo ...
194
195		return 0;
196	};
197
198Below is a matrix detailing which clk_ops are mandatory based upon the
199hardware capabilities of that clock.  A cell marked as "y" means
200mandatory, a cell marked as "n" implies that either including that
201callback is invalid or otherwise unnecessary.  Empty cells are either
202optional or must be evaluated on a case-by-case basis.
203
204.. table:: clock hardware characteristics
205
206   +----------------+------+-------------+---------------+-------------+------+
207   |                | gate | change rate | single parent | multiplexer | root |
208   +================+======+=============+===============+=============+======+
209   |.prepare        |      |             |               |             |      |
210   +----------------+------+-------------+---------------+-------------+------+
211   |.unprepare      |      |             |               |             |      |
212   +----------------+------+-------------+---------------+-------------+------+
213   +----------------+------+-------------+---------------+-------------+------+
214   |.enable         | y    |             |               |             |      |
215   +----------------+------+-------------+---------------+-------------+------+
216   |.disable        | y    |             |               |             |      |
217   +----------------+------+-------------+---------------+-------------+------+
218   |.is_enabled     | y    |             |               |             |      |
219   +----------------+------+-------------+---------------+-------------+------+
220   +----------------+------+-------------+---------------+-------------+------+
221   |.recalc_rate    |      | y           |               |             |      |
222   +----------------+------+-------------+---------------+-------------+------+
223   |.round_rate     |      | y [1]_      |               |             |      |
224   +----------------+------+-------------+---------------+-------------+------+
225   |.determine_rate |      | y [1]_      |               |             |      |
226   +----------------+------+-------------+---------------+-------------+------+
227   |.set_rate       |      | y           |               |             |      |
228   +----------------+------+-------------+---------------+-------------+------+
229   +----------------+------+-------------+---------------+-------------+------+
230   |.set_parent     |      |             | n             | y           | n    |
231   +----------------+------+-------------+---------------+-------------+------+
232   |.get_parent     |      |             | n             | y           | n    |
233   +----------------+------+-------------+---------------+-------------+------+
234   +----------------+------+-------------+---------------+-------------+------+
235   |.recalc_accuracy|      |             |               |             |      |
236   +----------------+------+-------------+---------------+-------------+------+
237   +----------------+------+-------------+---------------+-------------+------+
238   |.init           |      |             |               |             |      |
239   +----------------+------+-------------+---------------+-------------+------+
240
241.. [1] either one of round_rate or determine_rate is required.
242
243Finally, register your clock at run-time with a hardware-specific
244registration function.  This function simply populates struct clk_foo's
245data and then passes the common struct clk parameters to the framework
246with a call to::
247
248	clk_register(...)
249
250See the basic clock types in ``drivers/clk/clk-*.c`` for examples.
251
252Disabling clock gating of unused clocks
253=======================================
254
255Sometimes during development it can be useful to be able to bypass the
256default disabling of unused clocks. For example, if drivers aren't enabling
257clocks properly but rely on them being on from the bootloader, bypassing
258the disabling means that the driver will remain functional while the issues
259are sorted out.
260
261You can see which clocks have been disabled by booting your kernel with these
262parameters::
263
264 tp_printk trace_event=clk:clk_disable
265
266To bypass this disabling, include "clk_ignore_unused" in the bootargs to the
267kernel.
268
269Locking
270=======
271
272The common clock framework uses two global locks, the prepare lock and the
273enable lock.
274
275The enable lock is a spinlock and is held across calls to the .enable,
276.disable operations. Those operations are thus not allowed to sleep,
277and calls to the clk_enable(), clk_disable() API functions are allowed in
278atomic context.
279
280For clk_is_enabled() API, it is also designed to be allowed to be used in
281atomic context. However, it doesn't really make any sense to hold the enable
282lock in core, unless you want to do something else with the information of
283the enable state with that lock held. Otherwise, seeing if a clk is enabled is
284a one-shot read of the enabled state, which could just as easily change after
285the function returns because the lock is released. Thus the user of this API
286needs to handle synchronizing the read of the state with whatever they're
287using it for to make sure that the enable state doesn't change during that
288time.
289
290The prepare lock is a mutex and is held across calls to all other operations.
291All those operations are allowed to sleep, and calls to the corresponding API
292functions are not allowed in atomic context.
293
294This effectively divides operations in two groups from a locking perspective.
295
296Drivers don't need to manually protect resources shared between the operations
297of one group, regardless of whether those resources are shared by multiple
298clocks or not. However, access to resources that are shared between operations
299of the two groups needs to be protected by the drivers. An example of such a
300resource would be a register that controls both the clock rate and the clock
301enable/disable state.
302
303The clock framework is reentrant, in that a driver is allowed to call clock
304framework functions from within its implementation of clock operations. This
305can for instance cause a .set_rate operation of one clock being called from
306within the .set_rate operation of another clock. This case must be considered
307in the driver implementations, but the code flow is usually controlled by the
308driver in that case.
309
310Note that locking must also be considered when code outside of the common
311clock framework needs to access resources used by the clock operations. This
312is considered out of scope of this document.
313