Some OMAP2/3 hardware modules have CM_IDLEST attributes that are not
handled by the current omap2_wait_clock_ready() code. In preparation
for patches that fix the unusual devices, rename the function
omap2_wait_clock_ready() to omap2_wait_module_ready() and split it
into three parts:
1. A clkops-specific companion clock return function (by default,
omap2_clk_dflt_find_companion())
2. A clkops-specific CM_IDLEST register address and bit shift return
function (by default, omap2_clk_dflt_find_idlest())
3. Code to wait for the CM to indicate that the module is ready
(omap2_cm_wait_idlest())
Clocks can now specify their own custom find_companion() and find_idlest()
functions; used in subsequent patches.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
with while (i++ < MAX_CLOCK_ENABLE_WAIT); i can reach MAX_CLOCK_ENABLE_WAIT + 1
after the loop, so if (i == MAX_CLOCK_ENABLE_WAIT) that's still success.
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Remove OMAP_PRM_REGADDR and use processor specific defines instead.
Also fold in a patch from Kevin Hilman to add _OFFSET #defines
for the PRCM registers to be used with the prm_[read|write]_* macros.
These are used extensively in the forthcoming OMAP PM support.
Also remove now unused OMAP2_PRM_BASE.
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
On our system we see the following messages:
Disabling unused clock "gpt2_ick"
Disabling unused clock "gpt3_ick"
Disabling unused clock "gpt4_ick"
Disabling unused clock "gpt5_ick"
...
The messages have KERN_INFO level and if you have serial
console, they normally go there. I do not think it is good
idea to print that much stuff there. Moreover, messages
are not properly prefixed and for mortals it is not
immeadietly clear where they come from.
Let's give them debugging level instead.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Signed-off-by: Paul Walmsley <paul@pwsan.com>
[paul@pwsan.com: trimmed debugging output in patch description]
The CORE DPLL M2 frequency change code should use pr_debug(), not
pr_info(), for its debug messages. Same with
omap2_clksel_round_rate_div(). While here, convert a few printk(KERN_ERR ..
into pr_err().
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Richard Woodruff writes:
| The historic usage of this has been against single use leaf clocks
| (1st instance of gptimer). When it was used it did:
| clk_get()
| clk_set_parent()
| clk_enable()
|
| This usage was ok for that. Use on a disabled clock is needed.
|
| If there are multiple users on the clock or it is enabled there are
| problems.
|
| The call can still be unfriendly if 2 different drivers are using the
| clock with their own clock get/enable. It might be the function should
| return an error if usecount != 0 to stop surprises. It is all around
| better if the parenting is done when the clock is off.
This is a good reason to ensure that the clock is not enabled when
clk_set_parent() is called.
Acked-by: Richard Woodruff <r-woodruff2@ti.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This roughly corresponds with OMAP commits: 7d06c48, 3241b19,
88b5d9b, 18a5500, 9c909ac, 5c6497b, 8b1f0bd, 2ac1da8.
For both OMAP2 and OMAP3, we note the reference and bypass clocks in
the DPLL data structure. Whenever we modify the DPLL rate, we first
ensure that both the reference and bypass clocks are enabled. Then,
we decide whether to use the reference and DPLL, or the bypass clock
if the desired rate is identical to the bypass rate, and program the
DPLL appropriately. Finally, we update the clock's parent, and then
disable the unused clocks.
This keeps the parents correctly balanced, and more importantly ensures
that the bypass clock is running whenever we reprogram the DPLL. This
is especially important because the procedure for reprogramming the DPLL
involves switching to the bypass clock.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
When changing the parent of a clock, it is necessary to keep the
clock use counts balanced otherwise things the parent state will
get corrupted. Since we already disable and re-enable the clock,
we might as well use the recursive versions instead.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
_omap2_clksel_get_src_field() was returning the first entry which was
either the default _or_ applicable to the SoC. This is wrong - we
should be returning the first default which is applicable to the SoC.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The error checks for omap2_divisor_to_clksel() and comment disagree with
the actual value returned on error. Fix this to return the correct error
value.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
linux-omap source commit 33d000c99ee393fe2042f93e8422f94976d276ce
introduces a way to "dry run" clock changes before they're committed.
However, this involves putting logic to handle this into each and
every recalc function, and unfortunately due to the caching, led to
some bugs.
Solve both of issues by making the recalc methods always return the
clock rate for the clock, which the caller decides what to do with.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Based on a patch from Paul Walmsley <paul@pwsan.com>:
omap2_clk_enable() should enable a clock's clockdomain before
attempting to enable its parent clock's clockdomain. Similarly, in
the unlikely event that the parent clock enable fails, the clockdomain
should be disabled.
linux-omap source commit is 6d6e285e5a7912b1ea68fadac387304c914aaba8.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Based upon a patch from Paul Walmsley <paul@pwsan.com>:
If _omap2_clk_enable() fails, the clock's usecount must be decremented
by one no matter whether the clock has a parent or not.
but reorganised a bit.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Several parts of the OMAP2/3 clock code use wmb() to try to ensure
that the hardware write completes before continuing. This approach is
problematic: wmb() only ensures that the write leaves the ARM. It
does not ensure that the write actually reaches the endpoint device.
The endpoint device in this case - either the PRM, CM, or SCM - is
three interconnects away from the ARM - and the final interconnect is
low-speed. And the OCP interconnects will post the write, and who
knows how long that will take to complete. So the wmb() is not what
we want. Worse, the wmb() is indiscriminate; it causes the ARM to
flush any other unrelated buffered writes and wait for the local
interconnect to acknowledge them - potentially very expensive.
Fix this by converting the wmb()s into readbacks of the same PRM/CM/SCM
register. Since the PRM/CM/SCM devices use a single OCP thread, this
will cause the MPU to block while waiting for posted writes to that device
to complete.
linux-omap source commit is 260f5487848681b4d8ea7430a709a601bbcb21d1.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Consolidate the commit code for DELAYED_APP clocks into a subroutine,
_omap2xxx_clk_commit(). Also convert the MPU barrier wmb() into an
OCP barrier, since with an MPU barrier, we have no guarantee that the
write actually reached the endpoint device.
linux-omap source commit is 0f5bdb736515801b296125d16937a21ff7b3cfdc.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
clk_disable() previously used an ARM barrier, wmb(), to try to ensure
that the hardware write completed before continuing. There are some
problems with this approach.
The first problem is that wmb() only ensures that the write leaves the
ARM -- not that it actually reaches the endpoint device. In this
case, the endpoint device - either the PRM, CM, or SCM - is three
interconnects away from the ARM, and the final interconnect is
low-speed. And the OCP interconnects will post the write, who knows
how long that will take to complete. So the wmb() is not really what
we want.
Worse, the wmb() is indiscriminate; it will cause the ARM to flush any
other unrelated buffered writes and wait for the local interconnect to
acknowledge them - potentially very expensive.
This first problem could be fixed by doing a readback of the same PRM/CM/SCM
register. Since these devices use a single OCP thread, this will cause the
MPU to wait for the write to complete.
But the primary problem is a conceptual one: clk_disable() should not
need any kind of barrier. clk_enable() needs one since device driver
code must not access a device until its clocks are known to be
enabled. But clk_disable() has no such restriction.
Since blocking the MPU on a PRM/CM/SCM write can be a very
high-latency operation - several hundred MPU cycles - it's worth
avoiding this barrier if possible.
linux-omap source commit is f4aacad2c0ed1055622d5c1e910befece24ef0e2.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Traditionally, we've tracked the parent/child relationships between
clk structures by setting the child's parent member to point at the
upstream clock. As a result, when decending the tree, we have had
to scan all clocks to find the children.
Avoid this wasteful scanning by keeping a list of the clock's children.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Current implementation will disable clocks in the order defined in clock34xx.h,
at least DPLL4_M2X2 will hang in certain cases (and prevent retention / off)
if clocks are not disabled in correct order. This patch makes sure the parent
clocks will be active when disabling a clock.
linux-omap source commit is 672680063420ef8c8c4e7271984bb9cc08171d29.
Signed-off-by: Tero Kristo <tero.kristo@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Move the contents of the arch/arm/mach-omap2/memory.h file to the
existing mach/sdrc.h file, and remove memory.h. Modify files which
include memory.h to include asm/arch/sdrc.h instead.
linux-omap source commit is e7ae2d89921372fc4b9712a32cc401d645597807.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This patch rolls up several cleanup patches.
1. Some unnecessarily verbose variable names are used in several clock.c
functions; clean these up per CodingStyle.
2. Remove omap2_get_clksel() and just use clk->clksel_reg and
clk->clksel_mask directly.
3. Get rid of void __iomem * usage in omap2_clksel_get_src_field.
Prepend the function name with an underscore to highlight that it is a
static function.
linux-omap source commits are 7fa95e007ea2f3c4d0ecd2779d809756e7775894,
af0ea23f1ee4a5bea3b026e38761b47089f9048a, and
91c0c979b47c44b08f80e4f8d4c990fb158d82c4.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The DPLL FREQSEL jitter correction bits are set based on a table in
the 34xx TRM, Table 4-38, according to the DPLL's internal clock
frequency "Fint." Several Fint frequency ranges are missing from this
table. Previously, we allowed these Fint frequency ranges to be
selected in the rate rounding code, but did not change the FREQSEL bits.
Correspondence with the OMAP hardware team indicates that Fint values
not in the table should not be used. So, prevent them from being
selected during DPLL rate rounding. This removes warnings and also
can prevent the chip from locking up.
The first pass through the rate rounding code will update the DPLL max
and min dividers appropriately, so later rate rounding passes will run
faster than the first.
Peter de Schrijver <peter.de-schrijver@nokia.com> put up with several
test cycles of this patch - thanks Peter.
linux-omap source commit is f9c1b82f55b60fc39eaa6e7aa1fbe380c0ffe2e9.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Cc: Peter de Schrijver <peter.de-schrijver@nokia.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The previous DPLL rate rounding algorithm counted the divider (N) down
from the maximum to 1. Since we currently use a broad DPLL rate
tolerance, and lower N values are more power-efficient, we can often
bypass several iterations through the loop by counting N upwards from
1.
Peter de Schrijver <peter.de-schrijver@nokia.com> put up with several
test cycles of this patch - thanks Peter.
linux-omap source commit is 6f6d82bb2f80fa20a841ac3e95a6f44a5a156188.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Cc: Peter de Schrijver <peter.de-schrijver@nokia.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Remove some clutter from omap2_dpll_round_rate().
linux-omap source commit is 4625dceb8583c02a6d67ededc9f6a8347b6b8cb7.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Fix sparse & checkpatch warnings in OMAP2/3 PRCM & PM code. This mostly
consists of:
- converting pointer comparisons to integers in form similar to
(ptr == 0) to the standard idiom (!ptr)
- labeling a few non-static private functions as static
- adding prototypes for *_init() functions in the appropriate header
files, and getting rid of the corresponding open-coded extern
prototypes in other C files
- renaming the variable 'sclk' in mach-omap2/clock.c:omap2_get_apll_clkin
to avoid shadowing an earlier declaration
Clean up checkpatch issues. This mostly involves:
- converting some asm/ includes to linux/ includes
- cleaning up some whitespace
- getting rid of braces for conditionals with single following statements
Also take care of a few odds and ends, including:
- getting rid of unlikely() and likely() - none of this code is particularly
fast-path code, so the performance impact seems slim; and some of those
likely() and unlikely() indicators are probably not as accurate as the
ARM's branch predictor
- removing some superfluous casts
linux-omap source commit is 347df59f5d20fdf905afbc26b1328b0e28a8a01b.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
propagate_rate() is recursive, so it makes sense to minimise the
amount of stack which is used for each recursion. So, rather than
recursing back into it from the ->recalc functions if RATE_PROPAGATES
is set, do that test at the higher level.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Move the clock propagation calls for set_parent and set_rate into
the core omap clock code, rather than having these calls scattered
throughout the OMAP1 and OMAP2 implementations.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The original code in omap2_clk_wait_ready() used to check the low 8
bits to determine whether they were within the FCLKEN or ICLKEN
registers. Specifically, the test is satisfied when these offsets
are used:
CM_FCLKEN, CM_FCLKEN1, CM_CLKEN, OMAP24XX_CM_FCLKEN2, CM_ICLKEN,
CM_ICLKEN1, CM_ICLKEN2, CM_ICLKEN3, OMAP24XX_CM_ICLKEN4
OMAP3430_CM_CLKEN_PLL, OMAP3430ES2_CM_CLKEN2
If one of these offsets isn't used, omap2_clk_wait_ready() merely
returns without doing anything. So we should use the non-wait clkops
version instead and eliminate that conditional.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Rather than employing run-time tests in omap2_clk_wait_ready() to
decide whether we need to wait for the clock to become ready, we
can set the .ops appropriately.
This change deals with the OMAP24xx and OMAP34xx conditionals only.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
PARENT_CONTROLS_CLOCK just makes enable/disable no-op, and is
functionally an alias for ALWAYS_ENABLED. This can be handled
in the same way, using clkops_null.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
... and use it for clocks which are ALWAYS_ENABLED. These clocks
use a non-NULL enable_reg pointer for other purposes (such as
selecting clock rates.)
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Collect up all the common enable/disable clock operation functions
into a separate operations structure.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This patch integrates the OMAP3 clock tree with the clockdomain code.
This patch:
- marks OMAP34xx clocks with their corresponding clockdomain.
- adds code to convert the clockdomain name to a clockdomain pointer in the
struct clk during clk_register().
- modifies OMAP2 clock usecounting to call into the clockdomain code
when clocks are enabled or disabled.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
This patch adds a new rate rounding algorithm for DPLL clocks on the
OMAP2/3 architecture.
For a desired DPLL target rate, there may be several
multiplier/divider (M, N) values which will generate a sufficiently
close rate. Lower N values result in greater power economy. However,
lower N values can cause the difference between the rounded rate and
the target rate ("rate error") to be larger than it would be with a
higher N. This can cause downstream devices to run more slowly than
they otherwise would.
This DPLL rate rounding algorithm:
- attempts to find the lowest possible N (DPLL divider) to reach the
target_rate (since, according to Richard Woodruff <r-woodruff@ti.com>,
lower N values save more power than higher N values).
- allows developers to set an upper bound on the error between the
rounded rate and the desired target rate ("rate tolerance"), so an
appropriate balance between rate fidelity and power savings can be
set. This maximum rate error tolerance is set via
omap2_set_dpll_rate_tolerance().
- never returns a rounded rate higher than the target rate.
The rate rounding algorithm caches the last rounded M, N, and rate
computation to avoid rounding the rate twice for each clk_set_rate()
call. (This patch does not yet implement set_rate for DPLLs; that
follows in a future patch.)
The algorithm trades execution speed for rate accuracy. It will find
the (M, N) set that results in the least rate error, within a
specified rate tolerance. It does this by evaluating each divider
setting - on OMAP3, this involves 128 steps. Another approach to DPLL
rate rounding would be to bail out as soon as a valid rate is found
within the rate tolerance, which would trade rate accuracy for
execution speed. Alternate implementations welcome.
This code is not yet used by the OMAP24XX DPLL clock, since it
is currently defined as a composite clock, fusing the DPLL M,N and the
M2 output divider. This patch also renames the existing OMAP24xx DPLL
programming functions to highlight that they program both the DPLL and
the DPLL's output multiplier.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
ssi_l4_ick should have PARENT_CONTROLS_CLOCK flag.
ST_SSI_STDBY bit in idlest register cannot be used in omap2_clk_wait_ready
Signed-off-by: Jouni Högander <jouni.hogander@nokia.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
These changes is the result of the discussion with Paul Walmsley.
His ideas are included into this patch.
Remove DPLL output divider handling from DPLLs and CLKOUTX2 clocks,
and place it into specific DPLL output divider clocks (e.g., dpll3_m2_clk).
omap2_get_dpll_rate() now returns the correct DPLL rate, as represented
by the DPLL's CLKOUT output. Also add MPU and IVA2 subsystem clocks, along
with high-frequency bypass support.
Add support for DPLLs function in locked and bypass clock modes.
Signed-off-by: Roman Tereshonkov <roman.tereshonkov@nokia.com>
Acked-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
This patch changes 24xx to use shared clock code and new register
access.
Note that patch adds some temporary OLD_CK defines to keep patch
more readable. These temporary defines will be removed in the next
patch. Also not all clocks are changed in this patch to limit the
size.
Also, the patch fixes few incorrect clock defines in clock24xx.h.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
This patch adds a common clock framework for 24xx and 34xx.
Note that this patch does not add it to Makefile until in
next patch. Some functions are modified from earlier 24xx
clock framework code.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
This patch moves clock.h to clock24xx.c to make room for
adding common clock code for 24xx and 34xx.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
This patch moves clock.h to clock24xx.h to make room for
adding common clock code for 24xx and 34xx.
Signed-off-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
