blob: 1fee72f4d3319715f9efc2e258458cd807f3a537 [file] [log] [blame]
The Common Clk Framework
Mike Turquette <mturquette@ti.com>
This document endeavours to explain the common clk framework details,
and how to port a platform over to this framework. It is not yet a
detailed explanation of the clock api in include/linux/clk.h, but
perhaps someday it will include that information.
Part 1 - introduction and interface split
The common clk framework is an interface to control the clock nodes
available on various devices today. This may come in the form of clock
gating, rate adjustment, muxing or other operations. This framework is
enabled with the CONFIG_COMMON_CLK option.
The interface itself is divided into two halves, each shielded from the
details of its counterpart. First is the common definition of struct
clk which unifies the framework-level accounting and infrastructure that
has traditionally been duplicated across a variety of platforms. Second
is a common implementation of the clk.h api, defined in
drivers/clk/clk.c. Finally there is struct clk_ops, whose operations
are invoked by the clk api implementation.
The second half of the interface is comprised of the hardware-specific
callbacks registered with struct clk_ops and the corresponding
hardware-specific structures needed to model a particular clock. For
the remainder of this document any reference to a callback in struct
clk_ops, such as .enable or .set_rate, implies the hardware-specific
implementation of that code. Likewise, references to struct clk_foo
serve as a convenient shorthand for the implementation of the
hardware-specific bits for the hypothetical "foo" hardware.
Tying the two halves of this interface together is struct clk_hw, which
is defined in struct clk_foo and pointed to within struct clk. This
allows for easy navigation between the two discrete halves of the common
clock interface.
Part 2 - common data structures and api
Below is the common struct clk definition from
include/linux/clk-private.h, modified for brevity:
struct clk {
const char *name;
const struct clk_ops *ops;
struct clk_hw *hw;
char **parent_names;
struct clk **parents;
struct clk *parent;
struct hlist_head children;
struct hlist_node child_node;
...
};
The members above make up the core of the clk tree topology. The clk
api itself defines several driver-facing functions which operate on
struct clk. That api is documented in include/linux/clk.h.
Platforms and devices utilizing the common struct clk use the struct
clk_ops pointer in struct clk to perform the hardware-specific parts of
the operations defined in clk.h:
struct clk_ops {
int (*prepare)(struct clk_hw *hw);
void (*unprepare)(struct clk_hw *hw);
int (*enable)(struct clk_hw *hw);
void (*disable)(struct clk_hw *hw);
int (*is_enabled)(struct clk_hw *hw);
unsigned long (*recalc_rate)(struct clk_hw *hw,
unsigned long parent_rate);
long (*round_rate)(struct clk_hw *hw,
unsigned long rate,
unsigned long *parent_rate);
long (*determine_rate)(struct clk_hw *hw,
unsigned long rate,
unsigned long *best_parent_rate,
struct clk **best_parent_clk);
int (*set_parent)(struct clk_hw *hw, u8 index);
u8 (*get_parent)(struct clk_hw *hw);
int (*set_rate)(struct clk_hw *hw,
unsigned long rate,
unsigned long parent_rate);
int (*set_rate_and_parent)(struct clk_hw *hw,
unsigned long rate,
unsigned long parent_rate,
u8 index);
unsigned long (*recalc_accuracy)(struct clk_hw *hw,
unsigned long parent_accuracy);
void (*init)(struct clk_hw *hw);
int (*debug_init)(struct clk_hw *hw,
struct dentry *dentry);
};
Part 3 - hardware clk implementations
The strength of the common struct clk comes from its .ops and .hw pointers
which abstract the details of struct clk from the hardware-specific bits, and
vice versa. To illustrate consider the simple gateable clk implementation in
drivers/clk/clk-gate.c:
struct clk_gate {
struct clk_hw hw;
void __iomem *reg;
u8 bit_idx;
...
};
struct clk_gate contains struct clk_hw hw as well as hardware-specific
knowledge about which register and bit controls this clk's gating.
Nothing about clock topology or accounting, such as enable_count or
notifier_count, is needed here. That is all handled by the common
framework code and struct clk.
Let's walk through enabling this clk from driver code:
struct clk *clk;
clk = clk_get(NULL, "my_gateable_clk");
clk_prepare(clk);
clk_enable(clk);
The call graph for clk_enable is very simple:
clk_enable(clk);
clk->ops->enable(clk->hw);
[resolves to...]
clk_gate_enable(hw);
[resolves struct clk gate with to_clk_gate(hw)]
clk_gate_set_bit(gate);
And the definition of clk_gate_set_bit:
static void clk_gate_set_bit(struct clk_gate *gate)
{
u32 reg;
reg = __raw_readl(gate->reg);
reg |= BIT(gate->bit_idx);
writel(reg, gate->reg);
}
Note that to_clk_gate is defined as:
#define to_clk_gate(_hw) container_of(_hw, struct clk_gate, clk)
This pattern of abstraction is used for every clock hardware
representation.
Part 4 - supporting your own clk hardware
When implementing support for a new type of clock it only necessary to
include the following header:
#include <linux/clk-provider.h>
include/linux/clk.h is included within that header and clk-private.h
must never be included from the code which implements the operations for
a clock. More on that below in Part 5.
To construct a clk hardware structure for your platform you must define
the following:
struct clk_foo {
struct clk_hw hw;
... hardware specific data goes here ...
};
To take advantage of your data you'll need to support valid operations
for your clk:
struct clk_ops clk_foo_ops {
.enable = &clk_foo_enable;
.disable = &clk_foo_disable;
};
Implement the above functions using container_of:
#define to_clk_foo(_hw) container_of(_hw, struct clk_foo, hw)
int clk_foo_enable(struct clk_hw *hw)
{
struct clk_foo *foo;
foo = to_clk_foo(hw);
... perform magic on foo ...
return 0;
};
Below is a matrix detailing which clk_ops are mandatory based upon the
hardware capabilities of that clock. A cell marked as "y" means
mandatory, a cell marked as "n" implies that either including that
callback is invalid or otherwise unnecessary. Empty cells are either
optional or must be evaluated on a case-by-case basis.
clock hardware characteristics
-----------------------------------------------------------
| gate | change rate | single parent | multiplexer | root |
|------|-------------|---------------|-------------|------|
.prepare | | | | | |
.unprepare | | | | | |
| | | | | |
.enable | y | | | | |
.disable | y | | | | |
.is_enabled | y | | | | |
| | | | | |
.recalc_rate | | y | | | |
.round_rate | | y [1] | | | |
.determine_rate | | y [1] | | | |
.set_rate | | y | | | |
| | | | | |
.set_parent | | | n | y | n |
.get_parent | | | n | y | n |
| | | | | |
.recalc_accuracy| | | | | |
| | | | | |
.init | | | | | |
-----------------------------------------------------------
[1] either one of round_rate or determine_rate is required.
Finally, register your clock at run-time with a hardware-specific
registration function. This function simply populates struct clk_foo's
data and then passes the common struct clk parameters to the framework
with a call to:
clk_register(...)
See the basic clock types in drivers/clk/clk-*.c for examples.
Part 5 - static initialization of clock data
For platforms with many clocks (often numbering into the hundreds) it
may be desirable to statically initialize some clock data. This
presents a problem since the definition of struct clk should be hidden
from everyone except for the clock core in drivers/clk/clk.c.
To get around this problem struct clk's definition is exposed in
include/linux/clk-private.h along with some macros for more easily
initializing instances of the basic clock types. These clocks must
still be initialized with the common clock framework via a call to
__clk_init.
clk-private.h must NEVER be included by code which implements struct
clk_ops callbacks, nor must it be included by any logic which pokes
around inside of struct clk at run-time. To do so is a layering
violation.
To better enforce this policy, always follow this simple rule: any
statically initialized clock data MUST be defined in a separate file
from the logic that implements its ops. Basically separate the logic
from the data and all is well.
Part 6 - Disabling clock gating of unused clocks
Sometimes during development it can be useful to be able to bypass the
default disabling of unused clocks. For example, if drivers aren't enabling
clocks properly but rely on them being on from the bootloader, bypassing
the disabling means that the driver will remain functional while the issues
are sorted out.
To bypass this disabling, include "clk_ignore_unused" in the bootargs to the
kernel.
Part 7 - Locking
The common clock framework uses two global locks, the prepare lock and the
enable lock.
The enable lock is a spinlock and is held across calls to the .enable,
.disable and .is_enabled operations. Those operations are thus not allowed to
sleep, and calls to the clk_enable(), clk_disable() and clk_is_enabled() API
functions are allowed in atomic context.
The prepare lock is a mutex and is held across calls to all other operations.
All those operations are allowed to sleep, and calls to the corresponding API
functions are not allowed in atomic context.
This effectively divides operations in two groups from a locking perspective.
Drivers don't need to manually protect resources shared between the operations
of one group, regardless of whether those resources are shared by multiple
clocks or not. However, access to resources that are shared between operations
of the two groups needs to be protected by the drivers. An example of such a
resource would be a register that controls both the clock rate and the clock
enable/disable state.
The clock framework is reentrant, in that a driver is allowed to call clock
framework functions from within its implementation of clock operations. This
can for instance cause a .set_rate operation of one clock being called from
within the .set_rate operation of another clock. This case must be considered
in the driver implementations, but the code flow is usually controlled by the
driver in that case.
Note that locking must also be considered when code outside of the common
clock framework needs to access resources used by the clock operations. This
is considered out of scope of this document.