[v14,6/6] clk: meson: a1: add Amlogic A1 Peripherals clock controller driver

Commit Message

Dmitry Rokosov April 26, 2023, 9:58 a.m. UTC

Introduce Peripherals clock controller for Amlogic A1 SoC family.

A1 SoC has four clock controllers on the board: PLL, Peripherals, CPU,
and Audio.
This patchset adds support for Amlogic A1 Peripherals clock driver and
allows to generate clocks for all A1 SoC peripheral IPs.

Signed-off-by: Jian Hu <[e-mail]>
Signed-off-by: Dmitry Rokosov <[e-mail]>
---
 drivers/clk/meson/Kconfig  |   10 +
 drivers/clk/meson/Makefile |    1 +
 drivers/clk/meson/a1.c     | 2273 ++++++++++++++++++++++++++++++++++++
 drivers/clk/meson/a1.h     |  114 ++
 4 files changed, 2398 insertions(+)
 create mode 100644 drivers/clk/meson/a1.c
 create mode 100644 drivers/clk/meson/a1.h

Comments

Martin Blumenstingl May 1, 2023, 7:06 p.m. UTC | #1

Hi Dmitry,

overall this looks pretty good.

+Cc Heiner

On Wed, Apr 26, 2023 at 11:58 AM Dmitry Rokosov
<[e-mail]> wrote:
[...]
> +static struct clk_regmap pwm_a_sel = {
> +       .data = &(struct clk_regmap_mux_data){
> +               .offset = PWM_CLK_AB_CTRL,
> +               .mask = 0x1,
> +               .shift = 9,
> +       },
> +       .hw.init = &(struct clk_init_data){
> +               .name = "pwm_a_sel",
> +               .ops = &clk_regmap_mux_ops,
> +               .parent_data = pwm_abcd_parents,
> +               .num_parents = ARRAY_SIZE(pwm_abcd_parents),
> +               /* For more information, please refer to rtc clock */
> +               .flags = CLK_SET_RATE_NO_REPARENT,
Heiner is working on a series that adds common clock support to the
PWM driver [0].
I think his plans for a next step are adding support for SoCs where
the PWM clocks are part of the peripheral clock controller (instead of
being part of the PWM controller registers).

Have you considered removing CLK_SET_RATE_PARENT from the &rtc clock
so downstream clocks won't change the rtc clock rate by accident?
Then we could drop the CLK_SET_RATE_NO_REPARENT flag from the PWM
clocks to allow them to pick the best available parent (whether that's
the rtc clock, xtal or sys_pll).
That said, it would require managing the CLKID_RTC_32K_SEL clock (or
it's parents) using assigned-clocks instead of doing so with the PWM
(and other) clocks. Whether this would cause problems: I'm not sure,
so I'm hoping that you can share some insights.


Best regards,
Martin


[0] https://lore.kernel.org/linux-amlogic/[e-mail]/

Dmitry Rokosov May 12, 2023, 2:06 p.m. UTC | #2

Hello Martin,

On Mon, May 01, 2023 at 09:06:24PM +0200, Martin Blumenstingl wrote:
> Hi Dmitry,
> 
> overall this looks pretty good.

Thank you! Please find my thoughts about RTC clock below.

> 
> +Cc Heiner
> 
> On Wed, Apr 26, 2023 at 11:58 AM Dmitry Rokosov
> <[e-mail]> wrote:
> [...]
> > +static struct clk_regmap pwm_a_sel = {
> > +       .data = &(struct clk_regmap_mux_data){
> > +               .offset = PWM_CLK_AB_CTRL,
> > +               .mask = 0x1,
> > +               .shift = 9,
> > +       },
> > +       .hw.init = &(struct clk_init_data){
> > +               .name = "pwm_a_sel",
> > +               .ops = &clk_regmap_mux_ops,
> > +               .parent_data = pwm_abcd_parents,
> > +               .num_parents = ARRAY_SIZE(pwm_abcd_parents),
> > +               /* For more information, please refer to rtc clock */
> > +               .flags = CLK_SET_RATE_NO_REPARENT,
> Heiner is working on a series that adds common clock support to the
> PWM driver [0].
> I think his plans for a next step are adding support for SoCs where
> the PWM clocks are part of the peripheral clock controller (instead of
> being part of the PWM controller registers).
> 

Yes, I'm keeping up with this review and staying informed. It's worth
noting that the peripheral clock driver already includes PWM clocks,
with an important remark about reparenting being switched off. It's
described below.

> Have you considered removing CLK_SET_RATE_PARENT from the &rtc clock
> so downstream clocks won't change the rtc clock rate by accident?
> Then we could drop the CLK_SET_RATE_NO_REPARENT flag from the PWM
> clocks to allow them to pick the best available parent (whether that's
> the rtc clock, xtal or sys_pll).
> That said, it would require managing the CLKID_RTC_32K_SEL clock (or
> it's parents) using assigned-clocks instead of doing so with the PWM
> (and other) clocks. Whether this would cause problems: I'm not sure,
> so I'm hoping that you can share some insights.
> 
> 

Allow me to share my thoughts on this matter. From my understanding,
Amlogic provides an RTC clock that is both accurate and power-effective
in achieving a 32KHz signal from an internal xtal of 24MHz. However,
this requires a complex RTC divider with four parameters (m1, m2, n1,
n2), as it cannot be accomplished with a single divider. Our team has
measured the RTC clock using an oscilloscope on the GEN CLK pin and
found that it provides a stable 32KHz signal with acceptable jitter. On
the other hand, other approaches, such as the PWM way, yield less stable
and less accurate 32KHz signals with greater jitter.

Additionally, the CCF determines the best ancestor based on how close
its rate is to the given one, based on arithmetic calculations. However,
we have independent knowledge that a certain clock would be better, with
less jitter and fewer intermediaries, which will likely improve energy
efficiency. Sadly, the CCF cannot take this into account.

Given the advantages of the RTC clock, we wish to be able to control the
RTC as a parent for specific leaf clocks. This is achievable with the
'assigned-clocks' feature of CCF OF, but it poses a significant
architectural problem. The 'assigned-clocks' node does not lock/pin the
parent, and a simple clk_set_rate() call can change the parent during
rate propagation. In my opinion, an ideal solution to this problem would
be an additional patch to the CCF core that provides this locking
capability. As a board DTS developer, I know which clock I want to use
as the parent and have a strong reason for doing so, and I do not wish
to open up my parent muxing to other drivers. But until the behavior of
'assigned-clocks' is not available, we will simply label all RTC
children with the CLK_SET_RATE_NO_REPARENT flag.

> Best regards,
> Martin
> 
> 
> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/

Martin Blumenstingl May 16, 2023, 9:10 p.m. UTC | #3

Hello Dmitry,

On Fri, May 12, 2023 at 4:06 PM Dmitry Rokosov <[e-mail]> wrote:
[...]
> > > +static struct clk_regmap pwm_a_sel = {
> > > +       .data = &(struct clk_regmap_mux_data){
> > > +               .offset = PWM_CLK_AB_CTRL,
> > > +               .mask = 0x1,
> > > +               .shift = 9,
> > > +       },
> > > +       .hw.init = &(struct clk_init_data){
> > > +               .name = "pwm_a_sel",
> > > +               .ops = &clk_regmap_mux_ops,
> > > +               .parent_data = pwm_abcd_parents,
> > > +               .num_parents = ARRAY_SIZE(pwm_abcd_parents),
> > > +               /* For more information, please refer to rtc clock */
> > > +               .flags = CLK_SET_RATE_NO_REPARENT,
> > Heiner is working on a series that adds common clock support to the
> > PWM driver [0].
> > I think his plans for a next step are adding support for SoCs where
> > the PWM clocks are part of the peripheral clock controller (instead of
> > being part of the PWM controller registers).
> >
>
> Yes, I'm keeping up with this review and staying informed. It's worth
> noting that the peripheral clock driver already includes PWM clocks,
> with an important remark about reparenting being switched off. It's
> described below.
Indeed, this is why this question came to my mind

> > Have you considered removing CLK_SET_RATE_PARENT from the &rtc clock
> > so downstream clocks won't change the rtc clock rate by accident?
> > Then we could drop the CLK_SET_RATE_NO_REPARENT flag from the PWM
> > clocks to allow them to pick the best available parent (whether that's
> > the rtc clock, xtal or sys_pll).
> > That said, it would require managing the CLKID_RTC_32K_SEL clock (or
> > it's parents) using assigned-clocks instead of doing so with the PWM
> > (and other) clocks. Whether this would cause problems: I'm not sure,
> > so I'm hoping that you can share some insights.
> >
> >
>
> Allow me to share my thoughts on this matter. From my understanding,
> Amlogic provides an RTC clock that is both accurate and power-effective
> in achieving a 32KHz signal from an internal xtal of 24MHz. However,
> this requires a complex RTC divider with four parameters (m1, m2, n1,
> n2), as it cannot be accomplished with a single divider. Our team has
> measured the RTC clock using an oscilloscope on the GEN CLK pin and
> found that it provides a stable 32KHz signal with acceptable jitter. On
> the other hand, other approaches, such as the PWM way, yield less stable
> and less accurate 32KHz signals with greater jitter.
This part is clear to me (we may have even chatted on IRC how to use
the GEN CLK output previously)

> Additionally, the CCF determines the best ancestor based on how close
> its rate is to the given one, based on arithmetic calculations. However,
> we have independent knowledge that a certain clock would be better, with
> less jitter and fewer intermediaries, which will likely improve energy
> efficiency. Sadly, the CCF cannot take this into account.
I agree that the implementation in CCF is fairly simple. There's ways
to trick it though: IIRC if there are multiple equally suitable clocks
it picks the first one. For me all of this has worked so far which is
what makes me curious in this case (not saying that anything is wrong
with your approach).

Do you have a (real world) example where the RTC clock should be
preferred over another clock?

I'm thinking about the following scenario.
PWM parents:
- XTAL: 24MHz
- sys: not sure - let's say 166.67MHz
- RTC: 32kHz

Then after that there's a divider and a gate.

Let's say the PWM controller needs a 1MHz clock: it can take that from
XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
and use the divider.
But let's say the PWM controller needs a 32kHz clock: CCF would
automatically pick the RTC clock.
So is your implementation there to cover let's say 1kHz where
mathematically 24MHz can be divided evenly to 1kHz (and thus should
not result in any jitter) but RTC gives better precision in the real
world (even though it's off by 24Hz)?

> Given the advantages of the RTC clock, we wish to be able to control the
> RTC as a parent for specific leaf clocks. This is achievable with the
> 'assigned-clocks' feature of CCF OF, but it poses a significant
> architectural problem. The 'assigned-clocks' node does not lock/pin the
> parent, and a simple clk_set_rate() call can change the parent during
> rate propagation.
Are you aware of clk_set_rate_exclusive() and clk_rate_exclusive_{get,put}()?
It locks a clock and all of its parents to a certain rate. Other
consumers are unable to change the rate unless the lock is released
again.

> In my opinion, an ideal solution to this problem would
> be an additional patch to the CCF core that provides this locking
> capability.As a board DTS developer, I know which clock I want to use
> as the parent and have a strong reason for doing so, and I do not wish
> to open up my parent muxing to other drivers. But until the behavior of
> 'assigned-clocks' is not available, we will simply label all RTC
> children with the CLK_SET_RATE_NO_REPARENT flag.

PS: while writing this reply I found
drivers/clk/sunxi-ng/ccu-sun6i-rtc.c which implements
clk_ops.recalc_accuracy
I'm not sure I understand this correctly but it seems that CCF is not
using that information when making the decision which parent to use.


Best regards,
Martin

Dmitry Rokosov May 17, 2023, 10:34 a.m. UTC | #4

Hello Martin,

Thank you for reply and for sharing your thoughts, appreciate it!
Please find my comments below.

On Tue, May 16, 2023 at 11:10:19PM +0200, Martin Blumenstingl wrote:
> Hello Dmitry,
> 
> On Fri, May 12, 2023 at 4:06 PM Dmitry Rokosov <[e-mail]> wrote:
> [...]
> > > > +static struct clk_regmap pwm_a_sel = {
> > > > +       .data = &(struct clk_regmap_mux_data){
> > > > +               .offset = PWM_CLK_AB_CTRL,
> > > > +               .mask = 0x1,
> > > > +               .shift = 9,
> > > > +       },
> > > > +       .hw.init = &(struct clk_init_data){
> > > > +               .name = "pwm_a_sel",
> > > > +               .ops = &clk_regmap_mux_ops,
> > > > +               .parent_data = pwm_abcd_parents,
> > > > +               .num_parents = ARRAY_SIZE(pwm_abcd_parents),
> > > > +               /* For more information, please refer to rtc clock */
> > > > +               .flags = CLK_SET_RATE_NO_REPARENT,
> > > Heiner is working on a series that adds common clock support to the
> > > PWM driver [0].
> > > I think his plans for a next step are adding support for SoCs where
> > > the PWM clocks are part of the peripheral clock controller (instead of
> > > being part of the PWM controller registers).
> > >
> >
> > Yes, I'm keeping up with this review and staying informed. It's worth
> > noting that the peripheral clock driver already includes PWM clocks,
> > with an important remark about reparenting being switched off. It's
> > described below.
> Indeed, this is why this question came to my mind
> 
> > > Have you considered removing CLK_SET_RATE_PARENT from the &rtc clock
> > > so downstream clocks won't change the rtc clock rate by accident?
> > > Then we could drop the CLK_SET_RATE_NO_REPARENT flag from the PWM
> > > clocks to allow them to pick the best available parent (whether that's
> > > the rtc clock, xtal or sys_pll).
> > > That said, it would require managing the CLKID_RTC_32K_SEL clock (or
> > > it's parents) using assigned-clocks instead of doing so with the PWM
> > > (and other) clocks. Whether this would cause problems: I'm not sure,
> > > so I'm hoping that you can share some insights.
> > >
> > >
> >
> > Allow me to share my thoughts on this matter. From my understanding,
> > Amlogic provides an RTC clock that is both accurate and power-effective
> > in achieving a 32KHz signal from an internal xtal of 24MHz. However,
> > this requires a complex RTC divider with four parameters (m1, m2, n1,
> > n2), as it cannot be accomplished with a single divider. Our team has
> > measured the RTC clock using an oscilloscope on the GEN CLK pin and
> > found that it provides a stable 32KHz signal with acceptable jitter. On
> > the other hand, other approaches, such as the PWM way, yield less stable
> > and less accurate 32KHz signals with greater jitter.
> This part is clear to me (we may have even chatted on IRC how to use
> the GEN CLK output previously)
> 
> > Additionally, the CCF determines the best ancestor based on how close
> > its rate is to the given one, based on arithmetic calculations. However,
> > we have independent knowledge that a certain clock would be better, with
> > less jitter and fewer intermediaries, which will likely improve energy
> > efficiency. Sadly, the CCF cannot take this into account.
> I agree that the implementation in CCF is fairly simple. There's ways
> to trick it though: IIRC if there are multiple equally suitable clocks
> it picks the first one. For me all of this has worked so far which is
> what makes me curious in this case (not saying that anything is wrong
> with your approach).
> 
> Do you have a (real world) example where the RTC clock should be
> preferred over another clock?
> 

Yes, a real-life example is the need for a 32Khz clock for an external
wifi chip. There is one option to provide this clock with high
precision, which is RTC + GENCLK.

> I'm thinking about the following scenario.
> PWM parents:
> - XTAL: 24MHz
> - sys: not sure - let's say 166.67MHz
> - RTC: 32kHz
> 
> Then after that there's a divider and a gate.
> 
> Let's say the PWM controller needs a 1MHz clock: it can take that from
> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
> and use the divider.
> But let's say the PWM controller needs a 32kHz clock: CCF would
> automatically pick the RTC clock.
> So is your implementation there to cover let's say 1kHz where
> mathematically 24MHz can be divided evenly to 1kHz (and thus should
> not result in any jitter) but RTC gives better precision in the real
> world (even though it's off by 24Hz)?
> 

I don't think so. The highest precision that RTC can provide is from a
32KHz rate only. However, I believe that a 1kHz frequency can also be
achieved by using xtal 24MHz with a divider, which can provide high
precision as well.

> > Given the advantages of the RTC clock, we wish to be able to control the
> > RTC as a parent for specific leaf clocks. This is achievable with the
> > 'assigned-clocks' feature of CCF OF, but it poses a significant
> > architectural problem. The 'assigned-clocks' node does not lock/pin the
> > parent, and a simple clk_set_rate() call can change the parent during
> > rate propagation.
> Are you aware of clk_set_rate_exclusive() and clk_rate_exclusive_{get,put}()?
> It locks a clock and all of its parents to a certain rate. Other
> consumers are unable to change the rate unless the lock is released
> again.
> 

Agreed, this API works well for protecting the rate changes. However, I
do not think it covers the scenario where we want to change the rate but
still retain the parent (if it has already been assigned from the dts).

> > In my opinion, an ideal solution to this problem would
> > be an additional patch to the CCF core that provides this locking
> > capability.As a board DTS developer, I know which clock I want to use
> > as the parent and have a strong reason for doing so, and I do not wish
> > to open up my parent muxing to other drivers. But until the behavior of
> > 'assigned-clocks' is not available, we will simply label all RTC
> > children with the CLK_SET_RATE_NO_REPARENT flag.
> 
> PS: while writing this reply I found
> drivers/clk/sunxi-ng/ccu-sun6i-rtc.c which implements
> clk_ops.recalc_accuracy
> I'm not sure I understand this correctly but it seems that CCF is not
> using that information when making the decision which parent to use.

Hmm, I couldn't find how accuracy is used in the logic. It seems that
only the clk_summary contains information on accuracy.

I actually do not like my approach very much. CCF is not very flexible,
yet at the same time it does not allow for strict rules to be set for
the clock. With my approach, each developer would have to set up
assigned clocks directly from the device tree. However, if we back all
clocks inherited from RTC to the reparenting process, we can lose
assigned clocks for the RTC state.

As I mentioned earlier, I believe the best solution is to change CCF to
provide a new 'assigned-clocks-exclusive' mechanism while protecting the
clock from parent changing. I can provide an RFC patch for this if you
do not mind.

However, I think it is necessary to move forward with that patch series
since many other meson-a1.dtsi changes depend on it. We can merge the
current driver implementation as is, and after the CCF has an exclusive
assigned-clocks mechanism, then I will rework this driver.
Alternatively, I can revert all PARENT flags back and change the logic,
if it becomes necessary in the future on upstream boards.

Martin Blumenstingl May 18, 2023, 8:04 p.m. UTC | #5

Hi Dmitry,

On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
<[e-mail]> wrote:
[...]
> > > Additionally, the CCF determines the best ancestor based on how close
> > > its rate is to the given one, based on arithmetic calculations. However,
> > > we have independent knowledge that a certain clock would be better, with
> > > less jitter and fewer intermediaries, which will likely improve energy
> > > efficiency. Sadly, the CCF cannot take this into account.
> > I agree that the implementation in CCF is fairly simple. There's ways
> > to trick it though: IIRC if there are multiple equally suitable clocks
> > it picks the first one. For me all of this has worked so far which is
> > what makes me curious in this case (not saying that anything is wrong
> > with your approach).
> >
> > Do you have a (real world) example where the RTC clock should be
> > preferred over another clock?
> >
>
> Yes, a real-life example is the need for a 32Khz clock for an external
> wifi chip. There is one option to provide this clock with high
> precision, which is RTC + GENCLK.
>
> > I'm thinking about the following scenario.
> > PWM parents:
> > - XTAL: 24MHz
> > - sys: not sure - let's say 166.67MHz
> > - RTC: 32kHz
> >
> > Then after that there's a divider and a gate.
> >
> > Let's say the PWM controller needs a 1MHz clock: it can take that from
> > XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
> > and use the divider.
> > But let's say the PWM controller needs a 32kHz clock: CCF would
> > automatically pick the RTC clock.
> > So is your implementation there to cover let's say 1kHz where
> > mathematically 24MHz can be divided evenly to 1kHz (and thus should
> > not result in any jitter) but RTC gives better precision in the real
> > world (even though it's off by 24Hz)?
> >
>
> I don't think so. The highest precision that RTC can provide is from a
> 32KHz rate only. However, I believe that a 1kHz frequency can also be
> achieved by using xtal 24MHz with a divider, which can provide high
> precision as well.
Thank you again for the great discussion on IRC today.
Here's my short summary so I don't forget before you'll follow up on this.

In general there's two known cases where the RTC clock needs to be used:
a) When using the GENCLK output of the SoC to output the 32kHz RTC
clock and connect that to an SDIO WiFi chip clock input (this seems
useful in my understanding because the RTC clock provides high
precision)
b) When using the PWM controller to output a 32kHz clock signal. In
this case my understanding is that using the RTC clock as input to the
PWM controller results in the best possible signal

The second case won't be supported with Heiner's patches [0] that use
CCF (common clock framework) in the PWM controller driver.
In this series the parent clock is calculated using:
  freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);

A 32kHz clock means a PWM period of 30518ns. So with the above
calculation the PWM driver is asking for a clock rate of >=2GHz.
We concluded that letting the common clock framework choose the best
possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
be a way forward.
But this means that the PWM controller driver must try to find the
best possible parent somehow. The easiest way we came up with
(pseudo-code):
  freq = NSEC_PER_SEC / period;
  fin_freq = clk_round_rate(channel->clk, freq);
  if (fin_freq != freq) {
    freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
    fin_freq = clk_round_rate(channel->clk, freq);
  }

The idea is: for a requested 32kHz signal the PWM period is 30518ns.
The updated logic would find that there's a matching clock input and
use that directly. If not: use the original logic as suggested by
Heiner.


Best regards,
Martin


[0] https://lore.kernel.org/linux-amlogic/[e-mail]/

Dmitry Rokosov May 19, 2023, 3:07 p.m. UTC | #6

On Thu, May 18, 2023 at 10:04:41PM +0200, Martin Blumenstingl wrote:
> Hi Dmitry,
> 
> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
> <[e-mail]> wrote:
> [...]
> > > > Additionally, the CCF determines the best ancestor based on how close
> > > > its rate is to the given one, based on arithmetic calculations. However,
> > > > we have independent knowledge that a certain clock would be better, with
> > > > less jitter and fewer intermediaries, which will likely improve energy
> > > > efficiency. Sadly, the CCF cannot take this into account.
> > > I agree that the implementation in CCF is fairly simple. There's ways
> > > to trick it though: IIRC if there are multiple equally suitable clocks
> > > it picks the first one. For me all of this has worked so far which is
> > > what makes me curious in this case (not saying that anything is wrong
> > > with your approach).
> > >
> > > Do you have a (real world) example where the RTC clock should be
> > > preferred over another clock?
> > >
> >
> > Yes, a real-life example is the need for a 32Khz clock for an external
> > wifi chip. There is one option to provide this clock with high
> > precision, which is RTC + GENCLK.
> >
> > > I'm thinking about the following scenario.
> > > PWM parents:
> > > - XTAL: 24MHz
> > > - sys: not sure - let's say 166.67MHz
> > > - RTC: 32kHz
> > >
> > > Then after that there's a divider and a gate.
> > >
> > > Let's say the PWM controller needs a 1MHz clock: it can take that from
> > > XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
> > > and use the divider.
> > > But let's say the PWM controller needs a 32kHz clock: CCF would
> > > automatically pick the RTC clock.
> > > So is your implementation there to cover let's say 1kHz where
> > > mathematically 24MHz can be divided evenly to 1kHz (and thus should
> > > not result in any jitter) but RTC gives better precision in the real
> > > world (even though it's off by 24Hz)?
> > >
> >
> > I don't think so. The highest precision that RTC can provide is from a
> > 32KHz rate only. However, I believe that a 1kHz frequency can also be
> > achieved by using xtal 24MHz with a divider, which can provide high
> > precision as well.
> Thank you again for the great discussion on IRC today.
> Here's my short summary so I don't forget before you'll follow up on this.
> 
> In general there's two known cases where the RTC clock needs to be used:
> a) When using the GENCLK output of the SoC to output the 32kHz RTC
> clock and connect that to an SDIO WiFi chip clock input (this seems
> useful in my understanding because the RTC clock provides high
> precision)
> b) When using the PWM controller to output a 32kHz clock signal. In
> this case my understanding is that using the RTC clock as input to the
> PWM controller results in the best possible signal
> 
> The second case won't be supported with Heiner's patches [0] that use
> CCF (common clock framework) in the PWM controller driver.
> In this series the parent clock is calculated using:
>   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> 
> A 32kHz clock means a PWM period of 30518ns. So with the above
> calculation the PWM driver is asking for a clock rate of >=2GHz.
> We concluded that letting the common clock framework choose the best
> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
> be a way forward.
> But this means that the PWM controller driver must try to find the
> best possible parent somehow. The easiest way we came up with
> (pseudo-code):
>   freq = NSEC_PER_SEC / period;
>   fin_freq = clk_round_rate(channel->clk, freq);
>   if (fin_freq != freq) {
>     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>     fin_freq = clk_round_rate(channel->clk, freq);
>   }
> 
> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
> The updated logic would find that there's a matching clock input and
> use that directly. If not: use the original logic as suggested by
> Heiner.
> 
> 
> Best regards,
> Martin
> 
> 
> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/

Thank you for the excellent follow-up! I will reply to Heiner's thread
with these comments. Let's continue this discussion further in the
Heiner patch series.

Heiner Kallweit May 19, 2023, 4:10 p.m. UTC | #7

On 18.05.2023 22:04, Martin Blumenstingl wrote:
> Hi Dmitry,
> 
> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
> <[e-mail]> wrote:
> [...]
>>>> Additionally, the CCF determines the best ancestor based on how close
>>>> its rate is to the given one, based on arithmetic calculations. However,
>>>> we have independent knowledge that a certain clock would be better, with
>>>> less jitter and fewer intermediaries, which will likely improve energy
>>>> efficiency. Sadly, the CCF cannot take this into account.
>>> I agree that the implementation in CCF is fairly simple. There's ways
>>> to trick it though: IIRC if there are multiple equally suitable clocks
>>> it picks the first one. For me all of this has worked so far which is
>>> what makes me curious in this case (not saying that anything is wrong
>>> with your approach).
>>>
>>> Do you have a (real world) example where the RTC clock should be
>>> preferred over another clock?
>>>
>>
>> Yes, a real-life example is the need for a 32Khz clock for an external
>> wifi chip. There is one option to provide this clock with high
>> precision, which is RTC + GENCLK.
>>
>>> I'm thinking about the following scenario.
>>> PWM parents:
>>> - XTAL: 24MHz
>>> - sys: not sure - let's say 166.67MHz
>>> - RTC: 32kHz
>>>
>>> Then after that there's a divider and a gate.
>>>
>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
>>> and use the divider.
>>> But let's say the PWM controller needs a 32kHz clock: CCF would
>>> automatically pick the RTC clock.
>>> So is your implementation there to cover let's say 1kHz where
>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
>>> not result in any jitter) but RTC gives better precision in the real
>>> world (even though it's off by 24Hz)?
>>>
>>
>> I don't think so. The highest precision that RTC can provide is from a
>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
>> achieved by using xtal 24MHz with a divider, which can provide high
>> precision as well.
> Thank you again for the great discussion on IRC today.
> Here's my short summary so I don't forget before you'll follow up on this.
> 
> In general there's two known cases where the RTC clock needs to be used:
> a) When using the GENCLK output of the SoC to output the 32kHz RTC
> clock and connect that to an SDIO WiFi chip clock input (this seems
> useful in my understanding because the RTC clock provides high
> precision)
> b) When using the PWM controller to output a 32kHz clock signal. In
> this case my understanding is that using the RTC clock as input to the
> PWM controller results in the best possible signal
> 
> The second case won't be supported with Heiner's patches [0] that use
> CCF (common clock framework) in the PWM controller driver.
> In this series the parent clock is calculated using:
>   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> 
> A 32kHz clock means a PWM period of 30518ns. So with the above

To be precise: 30517,578125ns
What means that the PWM framework can't say "I want 32768Hz",
but just "I want something being very close to 32768Hz".
So what you need is some simple heuristic to interpret the
PWM request -> "PWM requests 30518ns, but supposedly it wants
32768Hz"

NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
because it tries to find the next lower clock.

The SoC families I'm familiar with have fclkin2 as PWM parent.
That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
for period = 30518n.
What you're saying is that newer generations don't have PWM parents
>24MHz any longer?


> calculation the PWM driver is asking for a clock rate of >=2GHz.
> We concluded that letting the common clock framework choose the best
> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
> be a way forward.
> But this means that the PWM controller driver must try to find the
> best possible parent somehow. The easiest way we came up with
> (pseudo-code):
>   freq = NSEC_PER_SEC / period;
>   fin_freq = clk_round_rate(channel->clk, freq);
>   if (fin_freq != freq) {
>     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>     fin_freq = clk_round_rate(channel->clk, freq);
>   }
> 
> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
> The updated logic would find that there's a matching clock input and
> use that directly. If not: use the original logic as suggested by
> Heiner.
> 
> 
> Best regards,
> Martin
> 
> 
> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/

Dmitry Rokosov May 22, 2023, 1:44 p.m. UTC | #8

Heiner,

On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
> On 18.05.2023 22:04, Martin Blumenstingl wrote:
> > Hi Dmitry,
> > 
> > On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
> > <[e-mail]> wrote:
> > [...]
> >>>> Additionally, the CCF determines the best ancestor based on how close
> >>>> its rate is to the given one, based on arithmetic calculations. However,
> >>>> we have independent knowledge that a certain clock would be better, with
> >>>> less jitter and fewer intermediaries, which will likely improve energy
> >>>> efficiency. Sadly, the CCF cannot take this into account.
> >>> I agree that the implementation in CCF is fairly simple. There's ways
> >>> to trick it though: IIRC if there are multiple equally suitable clocks
> >>> it picks the first one. For me all of this has worked so far which is
> >>> what makes me curious in this case (not saying that anything is wrong
> >>> with your approach).
> >>>
> >>> Do you have a (real world) example where the RTC clock should be
> >>> preferred over another clock?
> >>>
> >>
> >> Yes, a real-life example is the need for a 32Khz clock for an external
> >> wifi chip. There is one option to provide this clock with high
> >> precision, which is RTC + GENCLK.
> >>
> >>> I'm thinking about the following scenario.
> >>> PWM parents:
> >>> - XTAL: 24MHz
> >>> - sys: not sure - let's say 166.67MHz
> >>> - RTC: 32kHz
> >>>
> >>> Then after that there's a divider and a gate.
> >>>
> >>> Let's say the PWM controller needs a 1MHz clock: it can take that from
> >>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
> >>> and use the divider.
> >>> But let's say the PWM controller needs a 32kHz clock: CCF would
> >>> automatically pick the RTC clock.
> >>> So is your implementation there to cover let's say 1kHz where
> >>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
> >>> not result in any jitter) but RTC gives better precision in the real
> >>> world (even though it's off by 24Hz)?
> >>>
> >>
> >> I don't think so. The highest precision that RTC can provide is from a
> >> 32KHz rate only. However, I believe that a 1kHz frequency can also be
> >> achieved by using xtal 24MHz with a divider, which can provide high
> >> precision as well.
> > Thank you again for the great discussion on IRC today.
> > Here's my short summary so I don't forget before you'll follow up on this.
> > 
> > In general there's two known cases where the RTC clock needs to be used:
> > a) When using the GENCLK output of the SoC to output the 32kHz RTC
> > clock and connect that to an SDIO WiFi chip clock input (this seems
> > useful in my understanding because the RTC clock provides high
> > precision)
> > b) When using the PWM controller to output a 32kHz clock signal. In
> > this case my understanding is that using the RTC clock as input to the
> > PWM controller results in the best possible signal
> > 
> > The second case won't be supported with Heiner's patches [0] that use
> > CCF (common clock framework) in the PWM controller driver.
> > In this series the parent clock is calculated using:
> >   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> > 
> > A 32kHz clock means a PWM period of 30518ns. So with the above
> 
> To be precise: 30517,578125ns
> What means that the PWM framework can't say "I want 32768Hz",
> but just "I want something being very close to 32768Hz".
> So what you need is some simple heuristic to interpret the
> PWM request -> "PWM requests 30518ns, but supposedly it wants
> 32768Hz"
> 
> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
> because it tries to find the next lower clock.
> 
> The SoC families I'm familiar with have fclkin2 as PWM parent.
> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
> for period = 30518n.
> What you're saying is that newer generations don't have PWM parents
> >24MHz any longer?

No, of course not. For example, a fixed PLL (with all fclk_divX
settings) has rates higher than 24MHz. However, we need to consider the
'heavy' background of such PWM.

However, we have a "lightweight" clkin (special rtc32k) with a rate of
32kHz that we could potentially use as an input to produce a 32kHz
output on the PWM lines. I don't see any reason why we should not
support such special cases.

> 
> 
> > calculation the PWM driver is asking for a clock rate of >=2GHz.
> > We concluded that letting the common clock framework choose the best
> > possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
> > be a way forward.
> > But this means that the PWM controller driver must try to find the
> > best possible parent somehow. The easiest way we came up with
> > (pseudo-code):
> >   freq = NSEC_PER_SEC / period;
> >   fin_freq = clk_round_rate(channel->clk, freq);
> >   if (fin_freq != freq) {
> >     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> >     fin_freq = clk_round_rate(channel->clk, freq);
> >   }
> > 
> > The idea is: for a requested 32kHz signal the PWM period is 30518ns.
> > The updated logic would find that there's a matching clock input and
> > use that directly. If not: use the original logic as suggested by
> > Heiner.
> > 
> > 
> > Best regards,
> > Martin
> > 
> > 
> > [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
>

Heiner Kallweit May 22, 2023, 8:35 p.m. UTC | #9

On 22.05.2023 15:44, Dmitry Rokosov wrote:
> Heiner,
> 
> On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
>> On 18.05.2023 22:04, Martin Blumenstingl wrote:
>>> Hi Dmitry,
>>>
>>> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
>>> <[e-mail]> wrote:
>>> [...]
>>>>>> Additionally, the CCF determines the best ancestor based on how close
>>>>>> its rate is to the given one, based on arithmetic calculations. However,
>>>>>> we have independent knowledge that a certain clock would be better, with
>>>>>> less jitter and fewer intermediaries, which will likely improve energy
>>>>>> efficiency. Sadly, the CCF cannot take this into account.
>>>>> I agree that the implementation in CCF is fairly simple. There's ways
>>>>> to trick it though: IIRC if there are multiple equally suitable clocks
>>>>> it picks the first one. For me all of this has worked so far which is
>>>>> what makes me curious in this case (not saying that anything is wrong
>>>>> with your approach).
>>>>>
>>>>> Do you have a (real world) example where the RTC clock should be
>>>>> preferred over another clock?
>>>>>
>>>>
>>>> Yes, a real-life example is the need for a 32Khz clock for an external
>>>> wifi chip. There is one option to provide this clock with high
>>>> precision, which is RTC + GENCLK.
>>>>
>>>>> I'm thinking about the following scenario.
>>>>> PWM parents:
>>>>> - XTAL: 24MHz
>>>>> - sys: not sure - let's say 166.67MHz
>>>>> - RTC: 32kHz
>>>>>
>>>>> Then after that there's a divider and a gate.
>>>>>
>>>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
>>>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
>>>>> and use the divider.
>>>>> But let's say the PWM controller needs a 32kHz clock: CCF would
>>>>> automatically pick the RTC clock.
>>>>> So is your implementation there to cover let's say 1kHz where
>>>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
>>>>> not result in any jitter) but RTC gives better precision in the real
>>>>> world (even though it's off by 24Hz)?
>>>>>
>>>>
>>>> I don't think so. The highest precision that RTC can provide is from a
>>>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
>>>> achieved by using xtal 24MHz with a divider, which can provide high
>>>> precision as well.
>>> Thank you again for the great discussion on IRC today.
>>> Here's my short summary so I don't forget before you'll follow up on this.
>>>
>>> In general there's two known cases where the RTC clock needs to be used:
>>> a) When using the GENCLK output of the SoC to output the 32kHz RTC
>>> clock and connect that to an SDIO WiFi chip clock input (this seems
>>> useful in my understanding because the RTC clock provides high
>>> precision)
>>> b) When using the PWM controller to output a 32kHz clock signal. In
>>> this case my understanding is that using the RTC clock as input to the
>>> PWM controller results in the best possible signal
>>>
>>> The second case won't be supported with Heiner's patches [0] that use
>>> CCF (common clock framework) in the PWM controller driver.
>>> In this series the parent clock is calculated using:
>>>   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>
>>> A 32kHz clock means a PWM period of 30518ns. So with the above
>>
>> To be precise: 30517,578125ns
>> What means that the PWM framework can't say "I want 32768Hz",
>> but just "I want something being very close to 32768Hz".
>> So what you need is some simple heuristic to interpret the
>> PWM request -> "PWM requests 30518ns, but supposedly it wants
>> 32768Hz"
>>
>> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
>> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
>> because it tries to find the next lower clock.
>>
>> The SoC families I'm familiar with have fclkin2 as PWM parent.
>> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
>> for period = 30518n.
>> What you're saying is that newer generations don't have PWM parents
>>> 24MHz any longer?
> 
> No, of course not. For example, a fixed PLL (with all fclk_divX
> settings) has rates higher than 24MHz. However, we need to consider the
> 'heavy' background of such PWM.
> 
> However, we have a "lightweight" clkin (special rtc32k) with a rate of
> 32kHz that we could potentially use as an input to produce a 32kHz
> output on the PWM lines. I don't see any reason why we should not
> support such special cases.
> 

Two more things to consider:
1. When wanting a 32kHz (well, 32768Hz) output with a 50% duty cycle,
   then we need hi=0 and lo=0 with a 64kHz input clock.
   See point 2 for an explanation of why 0 and not 1.
   Means we couldn't use the RTC input clock. Did you consider this?
   Or do I miss something?
2. Seems the PWM block internally increments hi and lo, except the
   constant_en bit is set on newer PWM block versions.
   For bigger cnt values the impact is negligible, but for very small
   values it's something we have to consider.
   This was one additional motivation for me to choose an input
   frequency that creates big cnt values.

>>
>>
>>> calculation the PWM driver is asking for a clock rate of >=2GHz.
>>> We concluded that letting the common clock framework choose the best
>>> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
>>> be a way forward.
>>> But this means that the PWM controller driver must try to find the
>>> best possible parent somehow. The easiest way we came up with
>>> (pseudo-code):
>>>   freq = NSEC_PER_SEC / period;
>>>   fin_freq = clk_round_rate(channel->clk, freq);
>>>   if (fin_freq != freq) {
>>>     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>     fin_freq = clk_round_rate(channel->clk, freq);
>>>   }
>>>
>>> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
>>> The updated logic would find that there's a matching clock input and
>>> use that directly. If not: use the original logic as suggested by
>>> Heiner.
>>>
>>>
>>> Best regards,
>>> Martin
>>>
>>>
>>> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
>>
>

George Stark May 23, 2023, 9:42 a.m. UTC | #10

On 5/22/23 23:36, Heiner Kallweit wrote:
> On 22.05.2023 15:44, Dmitry Rokosov wrote:
>> Heiner,
>>
>> On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
>>> On 18.05.2023 22:04, Martin Blumenstingl wrote:
>>>> Hi Dmitry,
>>>>
>>>> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
>>>> <[e-mail]> wrote:
>>>> [...]
>>>>>>> Additionally, the CCF determines the best ancestor based on how close
>>>>>>> its rate is to the given one, based on arithmetic calculations. However,
>>>>>>> we have independent knowledge that a certain clock would be better, with
>>>>>>> less jitter and fewer intermediaries, which will likely improve energy
>>>>>>> efficiency. Sadly, the CCF cannot take this into account.
>>>>>> I agree that the implementation in CCF is fairly simple. There's ways
>>>>>> to trick it though: IIRC if there are multiple equally suitable clocks
>>>>>> it picks the first one. For me all of this has worked so far which is
>>>>>> what makes me curious in this case (not saying that anything is wrong
>>>>>> with your approach).
>>>>>>
>>>>>> Do you have a (real world) example where the RTC clock should be
>>>>>> preferred over another clock?
>>>>>>
>>>>> Yes, a real-life example is the need for a 32Khz clock for an external
>>>>> wifi chip. There is one option to provide this clock with high
>>>>> precision, which is RTC + GENCLK.
>>>>>
>>>>>> I'm thinking about the following scenario.
>>>>>> PWM parents:
>>>>>> - XTAL: 24MHz
>>>>>> - sys: not sure - let's say 166.67MHz
>>>>>> - RTC: 32kHz
>>>>>>
>>>>>> Then after that there's a divider and a gate.
>>>>>>
>>>>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
>>>>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
>>>>>> and use the divider.
>>>>>> But let's say the PWM controller needs a 32kHz clock: CCF would
>>>>>> automatically pick the RTC clock.
>>>>>> So is your implementation there to cover let's say 1kHz where
>>>>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
>>>>>> not result in any jitter) but RTC gives better precision in the real
>>>>>> world (even though it's off by 24Hz)?
>>>>>>
>>>>> I don't think so. The highest precision that RTC can provide is from a
>>>>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
>>>>> achieved by using xtal 24MHz with a divider, which can provide high
>>>>> precision as well.
>>>> Thank you again for the great discussion on IRC today.
>>>> Here's my short summary so I don't forget before you'll follow up on this.
>>>>
>>>> In general there's two known cases where the RTC clock needs to be used:
>>>> a) When using the GENCLK output of the SoC to output the 32kHz RTC
>>>> clock and connect that to an SDIO WiFi chip clock input (this seems
>>>> useful in my understanding because the RTC clock provides high
>>>> precision)
>>>> b) When using the PWM controller to output a 32kHz clock signal. In
>>>> this case my understanding is that using the RTC clock as input to the
>>>> PWM controller results in the best possible signal
>>>>
>>>> The second case won't be supported with Heiner's patches [0] that use
>>>> CCF (common clock framework) in the PWM controller driver.
>>>> In this series the parent clock is calculated using:
>>>>    freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>>
>>>> A 32kHz clock means a PWM period of 30518ns. So with the above
>>> To be precise: 30517,578125ns
>>> What means that the PWM framework can't say "I want 32768Hz",
>>> but just "I want something being very close to 32768Hz".
>>> So what you need is some simple heuristic to interpret the
>>> PWM request -> "PWM requests 30518ns, but supposedly it wants
>>> 32768Hz"
>>>
>>> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
>>> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
>>> because it tries to find the next lower clock.
>>>
>>> The SoC families I'm familiar with have fclkin2 as PWM parent.
>>> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
>>> for period = 30518n.
>>> What you're saying is that newer generations don't have PWM parents
>>>> 24MHz any longer?
>> No, of course not. For example, a fixed PLL (with all fclk_divX
>> settings) has rates higher than 24MHz. However, we need to consider the
>> 'heavy' background of such PWM.
>>
>> However, we have a "lightweight" clkin (special rtc32k) with a rate of
>> 32kHz that we could potentially use as an input to produce a 32kHz
>> output on the PWM lines. I don't see any reason why we should not
>> support such special cases.
>>
> Two more things to consider:
> 1. When wanting a 32kHz (well, 32768Hz) output with a 50% duty cycle,
>     then we need hi=0 and lo=0 with a 64kHz input clock.
>     See point 2 for an explanation of why 0 and not 1.
>     Means we couldn't use the RTC input clock. Did you consider this?
>     Or do I miss something?
> 2. Seems the PWM block internally increments hi and lo, except the
>     constant_en bit is set on newer PWM block versions.
>     For bigger cnt values the impact is negligible, but for very small
>     values it's something we have to consider.
>     This was one additional motivation for me to choose an input
>     frequency that creates big cnt values.

Hello Heiner

1. yes, you're right. To get pwm output 32k clock parent clock should be 64k at least.
2. yes, you're right. We have the same vision of pwm IP working. Seems like lo and hi regs are treated like divider regs (internal increment) except special case when constant bit on.

We have patches supporting constant bit andtaking into account hi\lo "shadow" incrementing, will public it soon
Best regards
George

>>>
>>>> calculation the PWM driver is asking for a clock rate of >=2GHz.
>>>> We concluded that letting the common clock framework choose the best
>>>> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
>>>> be a way forward.
>>>> But this means that the PWM controller driver must try to find the
>>>> best possible parent somehow. The easiest way we came up with
>>>> (pseudo-code):
>>>>    freq = NSEC_PER_SEC / period;
>>>>    fin_freq = clk_round_rate(channel->clk, freq);
>>>>    if (fin_freq != freq) {
>>>>      freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>>      fin_freq = clk_round_rate(channel->clk, freq);
>>>>    }
>>>>
>>>> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
>>>> The updated logic would find that there's a matching clock input and
>>>> use that directly. If not: use the original logic as suggested by
>>>> Heiner.
>>>>
>>>>
>>>> Best regards,
>>>> Martin
>>>>
>>>>
>>>> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
>

Dmitry Rokosov May 23, 2023, 10:23 a.m. UTC | #11

Heiner,

On Mon, May 22, 2023 at 10:35:59PM +0200, Heiner Kallweit wrote:
> On 22.05.2023 15:44, Dmitry Rokosov wrote:
> > Heiner,
> > 
> > On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
> >> On 18.05.2023 22:04, Martin Blumenstingl wrote:
> >>> Hi Dmitry,
> >>>
> >>> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
> >>> <[e-mail]> wrote:
> >>> [...]
> >>>>>> Additionally, the CCF determines the best ancestor based on how close
> >>>>>> its rate is to the given one, based on arithmetic calculations. However,
> >>>>>> we have independent knowledge that a certain clock would be better, with
> >>>>>> less jitter and fewer intermediaries, which will likely improve energy
> >>>>>> efficiency. Sadly, the CCF cannot take this into account.
> >>>>> I agree that the implementation in CCF is fairly simple. There's ways
> >>>>> to trick it though: IIRC if there are multiple equally suitable clocks
> >>>>> it picks the first one. For me all of this has worked so far which is
> >>>>> what makes me curious in this case (not saying that anything is wrong
> >>>>> with your approach).
> >>>>>
> >>>>> Do you have a (real world) example where the RTC clock should be
> >>>>> preferred over another clock?
> >>>>>
> >>>>
> >>>> Yes, a real-life example is the need for a 32Khz clock for an external
> >>>> wifi chip. There is one option to provide this clock with high
> >>>> precision, which is RTC + GENCLK.
> >>>>
> >>>>> I'm thinking about the following scenario.
> >>>>> PWM parents:
> >>>>> - XTAL: 24MHz
> >>>>> - sys: not sure - let's say 166.67MHz
> >>>>> - RTC: 32kHz
> >>>>>
> >>>>> Then after that there's a divider and a gate.
> >>>>>
> >>>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
> >>>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
> >>>>> and use the divider.
> >>>>> But let's say the PWM controller needs a 32kHz clock: CCF would
> >>>>> automatically pick the RTC clock.
> >>>>> So is your implementation there to cover let's say 1kHz where
> >>>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
> >>>>> not result in any jitter) but RTC gives better precision in the real
> >>>>> world (even though it's off by 24Hz)?
> >>>>>
> >>>>
> >>>> I don't think so. The highest precision that RTC can provide is from a
> >>>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
> >>>> achieved by using xtal 24MHz with a divider, which can provide high
> >>>> precision as well.
> >>> Thank you again for the great discussion on IRC today.
> >>> Here's my short summary so I don't forget before you'll follow up on this.
> >>>
> >>> In general there's two known cases where the RTC clock needs to be used:
> >>> a) When using the GENCLK output of the SoC to output the 32kHz RTC
> >>> clock and connect that to an SDIO WiFi chip clock input (this seems
> >>> useful in my understanding because the RTC clock provides high
> >>> precision)
> >>> b) When using the PWM controller to output a 32kHz clock signal. In
> >>> this case my understanding is that using the RTC clock as input to the
> >>> PWM controller results in the best possible signal
> >>>
> >>> The second case won't be supported with Heiner's patches [0] that use
> >>> CCF (common clock framework) in the PWM controller driver.
> >>> In this series the parent clock is calculated using:
> >>>   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> >>>
> >>> A 32kHz clock means a PWM period of 30518ns. So with the above
> >>
> >> To be precise: 30517,578125ns
> >> What means that the PWM framework can't say "I want 32768Hz",
> >> but just "I want something being very close to 32768Hz".
> >> So what you need is some simple heuristic to interpret the
> >> PWM request -> "PWM requests 30518ns, but supposedly it wants
> >> 32768Hz"
> >>
> >> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
> >> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
> >> because it tries to find the next lower clock.
> >>
> >> The SoC families I'm familiar with have fclkin2 as PWM parent.
> >> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
> >> for period = 30518n.
> >> What you're saying is that newer generations don't have PWM parents
> >>> 24MHz any longer?
> > 
> > No, of course not. For example, a fixed PLL (with all fclk_divX
> > settings) has rates higher than 24MHz. However, we need to consider the
> > 'heavy' background of such PWM.
> > 
> > However, we have a "lightweight" clkin (special rtc32k) with a rate of
> > 32kHz that we could potentially use as an input to produce a 32kHz
> > output on the PWM lines. I don't see any reason why we should not
> > support such special cases.
> > 
> 
> Two more things to consider:
> 1. When wanting a 32kHz (well, 32768Hz) output with a 50% duty cycle,
>    then we need hi=0 and lo=0 with a 64kHz input clock.
>    See point 2 for an explanation of why 0 and not 1.
>    Means we couldn't use the RTC input clock. Did you consider this?
>    Or do I miss something?

Nope, that's my fault. Using a 32kHz base clock for rates of 16kHz or
lower that can be divided evenly is a good choice. However, as you
mentioned in point 2, the problem with autoincrements should be fixed
first.

> 2. Seems the PWM block internally increments hi and lo, except the
>    constant_en bit is set on newer PWM block versions.
>    For bigger cnt values the impact is negligible, but for very small
>    values it's something we have to consider.
>    This was one additional motivation for me to choose an input
>    frequency that creates big cnt values.

George has a patchset for this and he will send it very soon as he
mentioned in another email. Let's assume the code is already fixed and
no longer has any issues. In my opinion, if we want to generate rates of
16kHz, 8kHz, 4kHz, etc., the best parent clock to use is a 32kHz RTC.
High-rate clocks may not divide evenly and may not generate an accurate
output rate. What do you think about it?

> 
> >>
> >>
> >>> calculation the PWM driver is asking for a clock rate of >=2GHz.
> >>> We concluded that letting the common clock framework choose the best
> >>> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
> >>> be a way forward.
> >>> But this means that the PWM controller driver must try to find the
> >>> best possible parent somehow. The easiest way we came up with
> >>> (pseudo-code):
> >>>   freq = NSEC_PER_SEC / period;
> >>>   fin_freq = clk_round_rate(channel->clk, freq);
> >>>   if (fin_freq != freq) {
> >>>     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> >>>     fin_freq = clk_round_rate(channel->clk, freq);
> >>>   }
> >>>
> >>> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
> >>> The updated logic would find that there's a matching clock input and
> >>> use that directly. If not: use the original logic as suggested by
> >>> Heiner.
> >>>
> >>>
> >>> Best regards,
> >>> Martin
> >>>
> >>>
> >>> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
> >>
> > 
>

Heiner Kallweit May 23, 2023, 7:13 p.m. UTC | #12

On 23.05.2023 12:23, Dmitry Rokosov wrote:
> Heiner,
> 
> On Mon, May 22, 2023 at 10:35:59PM +0200, Heiner Kallweit wrote:
>> On 22.05.2023 15:44, Dmitry Rokosov wrote:
>>> Heiner,
>>>
>>> On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
>>>> On 18.05.2023 22:04, Martin Blumenstingl wrote:
>>>>> Hi Dmitry,
>>>>>
>>>>> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
>>>>> <[e-mail]> wrote:
>>>>> [...]
>>>>>>>> Additionally, the CCF determines the best ancestor based on how close
>>>>>>>> its rate is to the given one, based on arithmetic calculations. However,
>>>>>>>> we have independent knowledge that a certain clock would be better, with
>>>>>>>> less jitter and fewer intermediaries, which will likely improve energy
>>>>>>>> efficiency. Sadly, the CCF cannot take this into account.
>>>>>>> I agree that the implementation in CCF is fairly simple. There's ways
>>>>>>> to trick it though: IIRC if there are multiple equally suitable clocks
>>>>>>> it picks the first one. For me all of this has worked so far which is
>>>>>>> what makes me curious in this case (not saying that anything is wrong
>>>>>>> with your approach).
>>>>>>>
>>>>>>> Do you have a (real world) example where the RTC clock should be
>>>>>>> preferred over another clock?
>>>>>>>
>>>>>>
>>>>>> Yes, a real-life example is the need for a 32Khz clock for an external
>>>>>> wifi chip. There is one option to provide this clock with high
>>>>>> precision, which is RTC + GENCLK.
>>>>>>
>>>>>>> I'm thinking about the following scenario.
>>>>>>> PWM parents:
>>>>>>> - XTAL: 24MHz
>>>>>>> - sys: not sure - let's say 166.67MHz
>>>>>>> - RTC: 32kHz
>>>>>>>
>>>>>>> Then after that there's a divider and a gate.
>>>>>>>
>>>>>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
>>>>>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
>>>>>>> and use the divider.
>>>>>>> But let's say the PWM controller needs a 32kHz clock: CCF would
>>>>>>> automatically pick the RTC clock.
>>>>>>> So is your implementation there to cover let's say 1kHz where
>>>>>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
>>>>>>> not result in any jitter) but RTC gives better precision in the real
>>>>>>> world (even though it's off by 24Hz)?
>>>>>>>
>>>>>>
>>>>>> I don't think so. The highest precision that RTC can provide is from a
>>>>>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
>>>>>> achieved by using xtal 24MHz with a divider, which can provide high
>>>>>> precision as well.
>>>>> Thank you again for the great discussion on IRC today.
>>>>> Here's my short summary so I don't forget before you'll follow up on this.
>>>>>
>>>>> In general there's two known cases where the RTC clock needs to be used:
>>>>> a) When using the GENCLK output of the SoC to output the 32kHz RTC
>>>>> clock and connect that to an SDIO WiFi chip clock input (this seems
>>>>> useful in my understanding because the RTC clock provides high
>>>>> precision)
>>>>> b) When using the PWM controller to output a 32kHz clock signal. In
>>>>> this case my understanding is that using the RTC clock as input to the
>>>>> PWM controller results in the best possible signal
>>>>>
>>>>> The second case won't be supported with Heiner's patches [0] that use
>>>>> CCF (common clock framework) in the PWM controller driver.
>>>>> In this series the parent clock is calculated using:
>>>>>   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>>>
>>>>> A 32kHz clock means a PWM period of 30518ns. So with the above
>>>>
>>>> To be precise: 30517,578125ns
>>>> What means that the PWM framework can't say "I want 32768Hz",
>>>> but just "I want something being very close to 32768Hz".
>>>> So what you need is some simple heuristic to interpret the
>>>> PWM request -> "PWM requests 30518ns, but supposedly it wants
>>>> 32768Hz"
>>>>
>>>> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
>>>> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
>>>> because it tries to find the next lower clock.
>>>>
>>>> The SoC families I'm familiar with have fclkin2 as PWM parent.
>>>> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
>>>> for period = 30518n.
>>>> What you're saying is that newer generations don't have PWM parents
>>>>> 24MHz any longer?
>>>
>>> No, of course not. For example, a fixed PLL (with all fclk_divX
>>> settings) has rates higher than 24MHz. However, we need to consider the
>>> 'heavy' background of such PWM.
>>>
>>> However, we have a "lightweight" clkin (special rtc32k) with a rate of
>>> 32kHz that we could potentially use as an input to produce a 32kHz
>>> output on the PWM lines. I don't see any reason why we should not
>>> support such special cases.
>>>
>>
>> Two more things to consider:
>> 1. When wanting a 32kHz (well, 32768Hz) output with a 50% duty cycle,
>>    then we need hi=0 and lo=0 with a 64kHz input clock.
>>    See point 2 for an explanation of why 0 and not 1.
>>    Means we couldn't use the RTC input clock. Did you consider this?
>>    Or do I miss something?
> 
> Nope, that's my fault. Using a 32kHz base clock for rates of 16kHz or
> lower that can be divided evenly is a good choice. However, as you
> mentioned in point 2, the problem with autoincrements should be fixed
> first.
> 
>> 2. Seems the PWM block internally increments hi and lo, except the
>>    constant_en bit is set on newer PWM block versions.
>>    For bigger cnt values the impact is negligible, but for very small
>>    values it's something we have to consider.
>>    This was one additional motivation for me to choose an input
>>    frequency that creates big cnt values.
> 
> George has a patchset for this and he will send it very soon as he
> mentioned in another email. Let's assume the code is already fixed and
> no longer has any issues. In my opinion, if we want to generate rates of
> 16kHz, 8kHz, 4kHz, etc., the best parent clock to use is a 32kHz RTC.
> High-rate clocks may not divide evenly and may not generate an accurate
> output rate. What do you think about it?
> 
With the logic in the CCF conversion patch set for these rates the 24MHz
xtal mux input would be used, creating exactly the requested rates.
So I see no difference here. What I don't know is whether power
consumption may be higher with a higher input rate, or whether quality
of the mux parent clocks differs.

>>
>>>>
>>>>
>>>>> calculation the PWM driver is asking for a clock rate of >=2GHz.
>>>>> We concluded that letting the common clock framework choose the best
>>>>> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
>>>>> be a way forward.
>>>>> But this means that the PWM controller driver must try to find the
>>>>> best possible parent somehow. The easiest way we came up with
>>>>> (pseudo-code):
>>>>>   freq = NSEC_PER_SEC / period;
>>>>>   fin_freq = clk_round_rate(channel->clk, freq);
>>>>>   if (fin_freq != freq) {
>>>>>     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>>>     fin_freq = clk_round_rate(channel->clk, freq);
>>>>>   }
>>>>>
>>>>> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
>>>>> The updated logic would find that there's a matching clock input and
>>>>> use that directly. If not: use the original logic as suggested by
>>>>> Heiner.
>>>>>
>>>>>
>>>>> Best regards,
>>>>> Martin
>>>>>
>>>>>
>>>>> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
>>>>
>>>
>>
>

George Stark May 24, 2023, 12:52 a.m. UTC | #13

On 5/22/23 23:36, Heiner Kallweit wrote:
> On 22.05.2023 15:44, Dmitry Rokosov wrote:
>> Heiner,
>>
>> On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
>>> On 18.05.2023 22:04, Martin Blumenstingl wrote:
>>>> Hi Dmitry,
>>>>
>>>> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
>>>> <[e-mail]> wrote:
>>>> [...]
>>>>>>> Additionally, the CCF determines the best ancestor based on how close
>>>>>>> its rate is to the given one, based on arithmetic calculations. However,
>>>>>>> we have independent knowledge that a certain clock would be better, with
>>>>>>> less jitter and fewer intermediaries, which will likely improve energy
>>>>>>> efficiency. Sadly, the CCF cannot take this into account.
>>>>>> I agree that the implementation in CCF is fairly simple. There's ways
>>>>>> to trick it though: IIRC if there are multiple equally suitable clocks
>>>>>> it picks the first one. For me all of this has worked so far which is
>>>>>> what makes me curious in this case (not saying that anything is wrong
>>>>>> with your approach).
>>>>>>
>>>>>> Do you have a (real world) example where the RTC clock should be
>>>>>> preferred over another clock?
>>>>>>
>>>>> Yes, a real-life example is the need for a 32Khz clock for an external
>>>>> wifi chip. There is one option to provide this clock with high
>>>>> precision, which is RTC + GENCLK.
>>>>>
>>>>>> I'm thinking about the following scenario.
>>>>>> PWM parents:
>>>>>> - XTAL: 24MHz
>>>>>> - sys: not sure - let's say 166.67MHz
>>>>>> - RTC: 32kHz
>>>>>>
>>>>>> Then after that there's a divider and a gate.
>>>>>>
>>>>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
>>>>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
>>>>>> and use the divider.
>>>>>> But let's say the PWM controller needs a 32kHz clock: CCF would
>>>>>> automatically pick the RTC clock.
>>>>>> So is your implementation there to cover let's say 1kHz where
>>>>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
>>>>>> not result in any jitter) but RTC gives better precision in the real
>>>>>> world (even though it's off by 24Hz)?
>>>>>>
>>>>> I don't think so. The highest precision that RTC can provide is from a
>>>>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
>>>>> achieved by using xtal 24MHz with a divider, which can provide high
>>>>> precision as well.
>>>> Thank you again for the great discussion on IRC today.
>>>> Here's my short summary so I don't forget before you'll follow up on this.
>>>>
>>>> In general there's two known cases where the RTC clock needs to be used:
>>>> a) When using the GENCLK output of the SoC to output the 32kHz RTC
>>>> clock and connect that to an SDIO WiFi chip clock input (this seems
>>>> useful in my understanding because the RTC clock provides high
>>>> precision)
>>>> b) When using the PWM controller to output a 32kHz clock signal. In
>>>> this case my understanding is that using the RTC clock as input to the
>>>> PWM controller results in the best possible signal
>>>>
>>>> The second case won't be supported with Heiner's patches [0] that use
>>>> CCF (common clock framework) in the PWM controller driver.
>>>> In this series the parent clock is calculated using:
>>>>    freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>>
>>>> A 32kHz clock means a PWM period of 30518ns. So with the above
>>> To be precise: 30517,578125ns
>>> What means that the PWM framework can't say "I want 32768Hz",
>>> but just "I want something being very close to 32768Hz".
>>> So what you need is some simple heuristic to interpret the
>>> PWM request -> "PWM requests 30518ns, but supposedly it wants
>>> 32768Hz"
>>>
>>> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
>>> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
>>> because it tries to find the next lower clock.
>>>
>>> The SoC families I'm familiar with have fclkin2 as PWM parent.
>>> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
>>> for period = 30518n.
>>> What you're saying is that newer generations don't have PWM parents
>>>> 24MHz any longer?
>> No, of course not. For example, a fixed PLL (with all fclk_divX
>> settings) has rates higher than 24MHz. However, we need to consider the
>> 'heavy' background of such PWM.
>>
>> However, we have a "lightweight" clkin (special rtc32k) with a rate of
>> 32kHz that we could potentially use as an input to produce a 32kHz
>> output on the PWM lines. I don't see any reason why we should not
>> support such special cases.
>>
> Two more things to consider:
> 1. When wanting a 32kHz (well, 32768Hz) output with a 50% duty cycle,
>     then we need hi=0 and lo=0 with a 64kHz input clock.
>     See point 2 for an explanation of why 0 and not 1.
>     Means we couldn't use the RTC input clock. Did you consider this?
>     Or do I miss something?
> 2. Seems the PWM block internally increments hi and lo, except the
>     constant_en bit is set on newer PWM block versions.
>     For bigger cnt values the impact is negligible, but for very small
>     values it's something we have to consider.
>     This was one additional motivation for me to choose an input
>     frequency that creates big cnt values.
>
Hello Heiner

As I mentioned earlier I have some changes to take into account lo and hi regs incrementing.

But it's more convenient to base my patch on top on one of yours (https://lore.kernel.org/linux-amlogic/[e-mail]/)

Is that ok if I resend your patch along with mine in series?

Best regards
George

>>>
>>>> calculation the PWM driver is asking for a clock rate of >=2GHz.
>>>> We concluded that letting the common clock framework choose the best
>>>> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
>>>> be a way forward.
>>>> But this means that the PWM controller driver must try to find the
>>>> best possible parent somehow. The easiest way we came up with
>>>> (pseudo-code):
>>>>    freq = NSEC_PER_SEC / period;
>>>>    fin_freq = clk_round_rate(channel->clk, freq);
>>>>    if (fin_freq != freq) {
>>>>      freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
>>>>      fin_freq = clk_round_rate(channel->clk, freq);
>>>>    }
>>>>
>>>> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
>>>> The updated logic would find that there's a matching clock input and
>>>> use that directly. If not: use the original logic as suggested by
>>>> Heiner.
>>>>
>>>>
>>>> Best regards,
>>>> Martin
>>>>
>>>>
>>>> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
>

Dmitry Rokosov May 24, 2023, 9:27 a.m. UTC | #14

On Tue, May 23, 2023 at 09:13:46PM +0200, Heiner Kallweit wrote:
> On 23.05.2023 12:23, Dmitry Rokosov wrote:
> > Heiner,
> > 
> > On Mon, May 22, 2023 at 10:35:59PM +0200, Heiner Kallweit wrote:
> >> On 22.05.2023 15:44, Dmitry Rokosov wrote:
> >>> Heiner,
> >>>
> >>> On Fri, May 19, 2023 at 06:10:50PM +0200, Heiner Kallweit wrote:
> >>>> On 18.05.2023 22:04, Martin Blumenstingl wrote:
> >>>>> Hi Dmitry,
> >>>>>
> >>>>> On Wed, May 17, 2023 at 12:34 PM Dmitry Rokosov
> >>>>> <[e-mail]> wrote:
> >>>>> [...]
> >>>>>>>> Additionally, the CCF determines the best ancestor based on how close
> >>>>>>>> its rate is to the given one, based on arithmetic calculations. However,
> >>>>>>>> we have independent knowledge that a certain clock would be better, with
> >>>>>>>> less jitter and fewer intermediaries, which will likely improve energy
> >>>>>>>> efficiency. Sadly, the CCF cannot take this into account.
> >>>>>>> I agree that the implementation in CCF is fairly simple. There's ways
> >>>>>>> to trick it though: IIRC if there are multiple equally suitable clocks
> >>>>>>> it picks the first one. For me all of this has worked so far which is
> >>>>>>> what makes me curious in this case (not saying that anything is wrong
> >>>>>>> with your approach).
> >>>>>>>
> >>>>>>> Do you have a (real world) example where the RTC clock should be
> >>>>>>> preferred over another clock?
> >>>>>>>
> >>>>>>
> >>>>>> Yes, a real-life example is the need for a 32Khz clock for an external
> >>>>>> wifi chip. There is one option to provide this clock with high
> >>>>>> precision, which is RTC + GENCLK.
> >>>>>>
> >>>>>>> I'm thinking about the following scenario.
> >>>>>>> PWM parents:
> >>>>>>> - XTAL: 24MHz
> >>>>>>> - sys: not sure - let's say 166.67MHz
> >>>>>>> - RTC: 32kHz
> >>>>>>>
> >>>>>>> Then after that there's a divider and a gate.
> >>>>>>>
> >>>>>>> Let's say the PWM controller needs a 1MHz clock: it can take that from
> >>>>>>> XTAL or sys. Since XTAL is evenly divisible to 1MHz CCF will pick that
> >>>>>>> and use the divider.
> >>>>>>> But let's say the PWM controller needs a 32kHz clock: CCF would
> >>>>>>> automatically pick the RTC clock.
> >>>>>>> So is your implementation there to cover let's say 1kHz where
> >>>>>>> mathematically 24MHz can be divided evenly to 1kHz (and thus should
> >>>>>>> not result in any jitter) but RTC gives better precision in the real
> >>>>>>> world (even though it's off by 24Hz)?
> >>>>>>>
> >>>>>>
> >>>>>> I don't think so. The highest precision that RTC can provide is from a
> >>>>>> 32KHz rate only. However, I believe that a 1kHz frequency can also be
> >>>>>> achieved by using xtal 24MHz with a divider, which can provide high
> >>>>>> precision as well.
> >>>>> Thank you again for the great discussion on IRC today.
> >>>>> Here's my short summary so I don't forget before you'll follow up on this.
> >>>>>
> >>>>> In general there's two known cases where the RTC clock needs to be used:
> >>>>> a) When using the GENCLK output of the SoC to output the 32kHz RTC
> >>>>> clock and connect that to an SDIO WiFi chip clock input (this seems
> >>>>> useful in my understanding because the RTC clock provides high
> >>>>> precision)
> >>>>> b) When using the PWM controller to output a 32kHz clock signal. In
> >>>>> this case my understanding is that using the RTC clock as input to the
> >>>>> PWM controller results in the best possible signal
> >>>>>
> >>>>> The second case won't be supported with Heiner's patches [0] that use
> >>>>> CCF (common clock framework) in the PWM controller driver.
> >>>>> In this series the parent clock is calculated using:
> >>>>>   freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> >>>>>
> >>>>> A 32kHz clock means a PWM period of 30518ns. So with the above
> >>>>
> >>>> To be precise: 30517,578125ns
> >>>> What means that the PWM framework can't say "I want 32768Hz",
> >>>> but just "I want something being very close to 32768Hz".
> >>>> So what you need is some simple heuristic to interpret the
> >>>> PWM request -> "PWM requests 30518ns, but supposedly it wants
> >>>> 32768Hz"
> >>>>
> >>>> NSEC_PER_SEC / 30518 = 32767 (rounded down from 32767,547)
> >>>> clk_round_rate(channel->clk, 32767) would return 0 (I *think*),
> >>>> because it tries to find the next lower clock.
> >>>>
> >>>> The SoC families I'm familiar with have fclkin2 as PWM parent.
> >>>> That's 1 GHz in my case, what results in a frequency of 32.767,547Hz
> >>>> for period = 30518n.
> >>>> What you're saying is that newer generations don't have PWM parents
> >>>>> 24MHz any longer?
> >>>
> >>> No, of course not. For example, a fixed PLL (with all fclk_divX
> >>> settings) has rates higher than 24MHz. However, we need to consider the
> >>> 'heavy' background of such PWM.
> >>>
> >>> However, we have a "lightweight" clkin (special rtc32k) with a rate of
> >>> 32kHz that we could potentially use as an input to produce a 32kHz
> >>> output on the PWM lines. I don't see any reason why we should not
> >>> support such special cases.
> >>>
> >>
> >> Two more things to consider:
> >> 1. When wanting a 32kHz (well, 32768Hz) output with a 50% duty cycle,
> >>    then we need hi=0 and lo=0 with a 64kHz input clock.
> >>    See point 2 for an explanation of why 0 and not 1.
> >>    Means we couldn't use the RTC input clock. Did you consider this?
> >>    Or do I miss something?
> > 
> > Nope, that's my fault. Using a 32kHz base clock for rates of 16kHz or
> > lower that can be divided evenly is a good choice. However, as you
> > mentioned in point 2, the problem with autoincrements should be fixed
> > first.
> > 
> >> 2. Seems the PWM block internally increments hi and lo, except the
> >>    constant_en bit is set on newer PWM block versions.
> >>    For bigger cnt values the impact is negligible, but for very small
> >>    values it's something we have to consider.
> >>    This was one additional motivation for me to choose an input
> >>    frequency that creates big cnt values.
> > 
> > George has a patchset for this and he will send it very soon as he
> > mentioned in another email. Let's assume the code is already fixed and
> > no longer has any issues. In my opinion, if we want to generate rates of
> > 16kHz, 8kHz, 4kHz, etc., the best parent clock to use is a 32kHz RTC.
> > High-rate clocks may not divide evenly and may not generate an accurate
> > output rate. What do you think about it?
> > 
> With the logic in the CCF conversion patch set for these rates the 24MHz
> xtal mux input would be used, creating exactly the requested rates.
> So I see no difference here. What I don't know is whether power
> consumption may be higher with a higher input rate, or whether quality
> of the mux parent clocks differs.
> 

We conducted some experiments on our end and tried using different
clocks as inputs for PWM to generate small rates like 16kHz. The clock
with the least amount of jitter for the resulting output was the RTC
parent at 32kHz, while using the 64MHz (system clock) or the 24MHz (xtal
clock) resulted in more jitter values but was still acceptable.

As for power consumption, I'm not knowledgeable about it; it's something
only Amlolgic has all the secrets to. However, I presume there was a
strong reason to create a separate RTC clock with good 32kHz quality.

> >>
> >>>>
> >>>>
> >>>>> calculation the PWM driver is asking for a clock rate of >=2GHz.
> >>>>> We concluded that letting the common clock framework choose the best
> >>>>> possible parent (meaning: removing CLK_SET_RATE_NO_REPARENT here) can
> >>>>> be a way forward.
> >>>>> But this means that the PWM controller driver must try to find the
> >>>>> best possible parent somehow. The easiest way we came up with
> >>>>> (pseudo-code):
> >>>>>   freq = NSEC_PER_SEC / period;
> >>>>>   fin_freq = clk_round_rate(channel->clk, freq);
> >>>>>   if (fin_freq != freq) {
> >>>>>     freq = div64_u64(NSEC_PER_SEC * (u64)0xffff, period);
> >>>>>     fin_freq = clk_round_rate(channel->clk, freq);
> >>>>>   }
> >>>>>
> >>>>> The idea is: for a requested 32kHz signal the PWM period is 30518ns.
> >>>>> The updated logic would find that there's a matching clock input and
> >>>>> use that directly. If not: use the original logic as suggested by
> >>>>> Heiner.
> >>>>>
> >>>>>
> >>>>> Best regards,
> >>>>> Martin
> >>>>>
> >>>>>
> >>>>> [0] https://lore.kernel.org/linux-amlogic/[e-mail]/
> >>>>
> >>>
> >>
> > 
>

Patch

diff --git a/drivers/clk/meson/Kconfig b/drivers/clk/meson/Kconfig
index 693c4cf60f27..dff576363b1f 100644
--- a/drivers/clk/meson/Kconfig
+++ b/drivers/clk/meson/Kconfig
@@ -109,6 +109,16 @@  config COMMON_CLK_A1_PLL
 	  device, A1 SoC Family. Say Y if you want A1 PLL clock controller
 	  to work.
 
+config COMMON_CLK_A1
+	tristate "Amlogic A1 SoC clock controller support"
+	depends on ARM64
+	select COMMON_CLK_MESON_DUALDIV
+	select COMMON_CLK_MESON_REGMAP
+	help
+	  Support for the Peripherals clock controller on Amlogic A113L based
+	  device, A1 SoC Family. Say Y if you want A1 Peripherals clock
+	  controller to work.
+
 config COMMON_CLK_G12A
 	tristate "G12 and SM1 SoC clock controllers support"
 	depends on ARM64
diff --git a/drivers/clk/meson/Makefile b/drivers/clk/meson/Makefile
index 2f17f475a48f..0e6f293c05d4 100644
--- a/drivers/clk/meson/Makefile
+++ b/drivers/clk/meson/Makefile
@@ -17,6 +17,7 @@  obj-$(CONFIG_COMMON_CLK_MESON_VID_PLL_DIV) += vid-pll-div.o
 obj-$(CONFIG_COMMON_CLK_AXG) += axg.o axg-aoclk.o
 obj-$(CONFIG_COMMON_CLK_AXG_AUDIO) += axg-audio.o
 obj-$(CONFIG_COMMON_CLK_A1_PLL) += a1-pll.o
+obj-$(CONFIG_COMMON_CLK_A1) += a1.o
 obj-$(CONFIG_COMMON_CLK_GXBB) += gxbb.o gxbb-aoclk.o
 obj-$(CONFIG_COMMON_CLK_G12A) += g12a.o g12a-aoclk.o
 obj-$(CONFIG_COMMON_CLK_MESON8B) += meson8b.o meson8-ddr.o
diff --git a/drivers/clk/meson/a1.c b/drivers/clk/meson/a1.c
new file mode 100644
index 000000000000..51b50b3ff100
--- /dev/null
+++ b/drivers/clk/meson/a1.c
@@ -0,0 +1,2273 @@ 
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright (c) 2019 Amlogic, Inc. All rights reserved.
+ * Author: Jian Hu <[e-mail]>
+ *
+ * Copyright (c) 2023, SberDevices. All Rights Reserved.
+ * Author: Dmitry Rokosov <[e-mail]>
+ */
+
+#include <linux/clk-provider.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include "a1.h"
+#include "clk-dualdiv.h"
+#include "clk-regmap.h"
+
+static struct clk_regmap xtal_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 0,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "xtal_in",
+		.ops = &clk_regmap_gate_ro_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap fixpll_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 1,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "fixpll_in",
+		.ops = &clk_regmap_gate_ro_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap usb_phy_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 2,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "usb_phy_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap usb_ctrl_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 3,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "usb_ctrl_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap hifipll_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 4,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "hifipll_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap syspll_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 5,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "syspll_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap dds_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_OSCIN_CTRL,
+		.bit_idx = 6,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dds_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap rtc_32k_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = RTC_BY_OSCIN_CTRL0,
+		.bit_idx = 31,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "rtc_32k_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static const struct meson_clk_dualdiv_param clk_32k_div_table[] = {
+	{
+		.dual		= 1,
+		.n1		= 733,
+		.m1		= 8,
+		.n2		= 732,
+		.m2		= 11,
+	},
+	{}
+};
+
+static struct clk_regmap rtc_32k_div = {
+	.data = &(struct meson_clk_dualdiv_data){
+		.n1 = {
+			.reg_off = RTC_BY_OSCIN_CTRL0,
+			.shift   = 0,
+			.width   = 12,
+		},
+		.n2 = {
+			.reg_off = RTC_BY_OSCIN_CTRL0,
+			.shift   = 12,
+			.width   = 12,
+		},
+		.m1 = {
+			.reg_off = RTC_BY_OSCIN_CTRL1,
+			.shift   = 0,
+			.width   = 12,
+		},
+		.m2 = {
+			.reg_off = RTC_BY_OSCIN_CTRL1,
+			.shift   = 12,
+			.width   = 12,
+		},
+		.dual = {
+			.reg_off = RTC_BY_OSCIN_CTRL0,
+			.shift   = 28,
+			.width   = 1,
+		},
+		.table = clk_32k_div_table,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "rtc_32k_div",
+		.ops = &meson_clk_dualdiv_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&rtc_32k_in.hw
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap rtc_32k_xtal = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = RTC_BY_OSCIN_CTRL1,
+		.bit_idx = 24,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "rtc_32k_xtal",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&rtc_32k_in.hw
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap rtc_32k_sel = {
+	.data = &(struct clk_regmap_mux_data) {
+		.offset = RTC_CTRL,
+		.mask = 0x3,
+		.shift = 0,
+		.flags = CLK_MUX_ROUND_CLOSEST,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "rtc_32k_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&rtc_32k_xtal.hw,
+			&rtc_32k_div.hw,
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+/*
+ * All clocks that can be inherited from a more accurate RTC clock are marked
+ * with the CLK_SET_RATE_NO_REPARENT flag. This is because in certain
+ * situations, we may need to freeze their parent. The parent setup of these
+ * clocks should be located on the device tree side.
+ */
+struct clk_regmap rtc = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = RTC_BY_OSCIN_CTRL0,
+		.bit_idx = 30,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "rtc",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&rtc_32k_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static u32 mux_table_sys[] = { 0, 1, 2, 3, 7 };
+static const struct clk_parent_data sys_parents[] = {
+	{ .fw_name = "xtal" },
+	{ .fw_name = "fclk_div2" },
+	{ .fw_name = "fclk_div3" },
+	{ .fw_name = "fclk_div5" },
+	{ .hw = &rtc.hw },
+};
+
+static struct clk_regmap sys_b_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SYS_CLK_CTRL0,
+		.mask = 0x7,
+		.shift = 26,
+		.table = mux_table_sys,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sys_b_sel",
+		.ops = &clk_regmap_mux_ro_ops,
+		.parent_data = sys_parents,
+		.num_parents = ARRAY_SIZE(sys_parents),
+	},
+};
+
+static struct clk_regmap sys_b_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = SYS_CLK_CTRL0,
+		.shift = 16,
+		.width = 10,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sys_b_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sys_b_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap sys_b = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_CLK_CTRL0,
+		.bit_idx = 29,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "sys_b",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sys_b_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap sys_a_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SYS_CLK_CTRL0,
+		.mask = 0x7,
+		.shift = 10,
+		.table = mux_table_sys,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sys_a_sel",
+		.ops = &clk_regmap_mux_ro_ops,
+		.parent_data = sys_parents,
+		.num_parents = ARRAY_SIZE(sys_parents),
+	},
+};
+
+static struct clk_regmap sys_a_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = SYS_CLK_CTRL0,
+		.shift = 0,
+		.width = 10,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sys_a_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sys_a_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap sys_a = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SYS_CLK_CTRL0,
+		.bit_idx = 13,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "sys_a",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sys_a_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap sys = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SYS_CLK_CTRL0,
+		.mask = 0x1,
+		.shift = 31,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sys",
+		.ops = &clk_regmap_mux_ro_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sys_a.hw,
+			&sys_b.hw,
+		},
+		.num_parents = 2,
+		/*
+		 * This clock is used by APB bus which is set in boot ROM code
+		 * and is required by the platform to operate correctly.
+		 * Until the following condition are met, we need this clock to
+		 * be marked as critical:
+		 * a) Mark the clock used by a firmware resource, if possible
+		 * b) CCF has a clock hand-off mechanism to make the sure the
+		 *    clock stays on until the proper driver comes along
+		 */
+		.flags = CLK_SET_RATE_PARENT | CLK_IS_CRITICAL,
+	},
+};
+
+static u32 mux_table_dsp_ab[] = { 0, 1, 2, 3, 4, 7 };
+static const struct clk_parent_data dsp_ab_parent_data[] = {
+	{ .fw_name = "xtal", },
+	{ .fw_name = "fclk_div2", },
+	{ .fw_name = "fclk_div3", },
+	{ .fw_name = "fclk_div5", },
+	{ .fw_name = "hifi_pll", },
+	{ .hw = &rtc.hw },
+};
+
+static struct clk_regmap dspa_a_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DSPA_CLK_CTRL0,
+		.mask = 0x7,
+		.shift = 10,
+		.table = mux_table_dsp_ab,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspa_a_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = dsp_ab_parent_data,
+		.num_parents = ARRAY_SIZE(dsp_ab_parent_data),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap dspa_a_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = DSPA_CLK_CTRL0,
+		.shift = 0,
+		.width = 10,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspa_a_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_a_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspa_a = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPA_CLK_CTRL0,
+		.bit_idx = 13,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspa_a",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_a_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspa_b_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DSPA_CLK_CTRL0,
+		.mask = 0x7,
+		.shift = 26,
+		.table = mux_table_dsp_ab,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspa_b_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = dsp_ab_parent_data,
+		.num_parents = ARRAY_SIZE(dsp_ab_parent_data),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap dspa_b_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = DSPA_CLK_CTRL0,
+		.shift = 16,
+		.width = 10,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspa_b_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_b_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspa_b = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPA_CLK_CTRL0,
+		.bit_idx = 29,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspa_b",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_b_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspa_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DSPA_CLK_CTRL0,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspa_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_a.hw,
+			&dspa_b.hw,
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspa_en = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPA_CLK_EN,
+		.bit_idx = 1,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspa_en",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspa_en_nic = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPA_CLK_EN,
+		.bit_idx = 0,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspa_en_nic",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspa_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_a_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DSPB_CLK_CTRL0,
+		.mask = 0x7,
+		.shift = 10,
+		.table = mux_table_dsp_ab,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspb_a_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = dsp_ab_parent_data,
+		.num_parents = ARRAY_SIZE(dsp_ab_parent_data),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap dspb_a_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = DSPB_CLK_CTRL0,
+		.shift = 0,
+		.width = 10,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspb_a_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_a_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_a = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPB_CLK_CTRL0,
+		.bit_idx = 13,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspb_a",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_a_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_b_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DSPB_CLK_CTRL0,
+		.mask = 0x7,
+		.shift = 26,
+		.table = mux_table_dsp_ab,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspb_b_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = dsp_ab_parent_data,
+		.num_parents = ARRAY_SIZE(dsp_ab_parent_data),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap dspb_b_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = DSPB_CLK_CTRL0,
+		.shift = 16,
+		.width = 10,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspb_b_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_b_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_b = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPB_CLK_CTRL0,
+		.bit_idx = 29,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspb_b",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_b_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DSPB_CLK_CTRL0,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dspb_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_a.hw,
+			&dspb_b.hw,
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_en = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPB_CLK_EN,
+		.bit_idx = 1,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspb_en",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dspb_en_nic = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DSPB_CLK_EN,
+		.bit_idx = 0,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dspb_en_nic",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dspb_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap clk_24m = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CLK12_24_CTRL,
+		.bit_idx = 11,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "24m",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_fixed_factor clk_24m_div2 = {
+	.mult = 1,
+	.div = 2,
+	.hw.init = &(struct clk_init_data){
+		.name = "24m_div2",
+		.ops = &clk_fixed_factor_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&clk_24m.hw
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap clk_12m = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CLK12_24_CTRL,
+		.bit_idx = 10,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "12m",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&clk_24m_div2.hw
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap fclk_div2_divn_pre = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = CLK12_24_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "fclk_div2_divn_pre",
+		.ops = &clk_regmap_divider_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "fclk_div2",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap fclk_div2_divn = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CLK12_24_CTRL,
+		.bit_idx = 12,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "fclk_div2_divn",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&fclk_div2_divn_pre.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+/*
+ * the index 2 is sys_pll_div16, it will be implemented in the CPU clock driver,
+ * the index 4 is the clock measurement source, it's not supported yet
+ */
+static u32 gen_table[] = { 0, 1, 3, 5, 6, 7, 8 };
+static const struct clk_parent_data gen_parent_data[] = {
+	{ .fw_name = "xtal", },
+	{ .hw = &rtc.hw },
+	{ .fw_name = "hifi_pll", },
+	{ .fw_name = "fclk_div2", },
+	{ .fw_name = "fclk_div3", },
+	{ .fw_name = "fclk_div5", },
+	{ .fw_name = "fclk_div7", },
+};
+
+static struct clk_regmap gen_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = GEN_CLK_CTRL,
+		.mask = 0xf,
+		.shift = 12,
+		.table = gen_table,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "gen_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = gen_parent_data,
+		.num_parents = ARRAY_SIZE(gen_parent_data),
+		/*
+		 * The GEN clock can be connected to an external pad, so it
+		 * may be set up directly from the device tree. Additionally,
+		 * the GEN clock can be inherited from a more accurate RTC
+		 * clock, so in certain situations, it may be necessary
+		 * to freeze its parent.
+		 */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap gen_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = GEN_CLK_CTRL,
+		.shift = 0,
+		.width = 11,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "gen_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&gen_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap gen = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = GEN_CLK_CTRL,
+		.bit_idx = 11,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "gen",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&gen_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap saradc_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SAR_ADC_CLK_CTRL,
+		.mask = 0x1,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "saradc_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = (const struct clk_parent_data []) {
+			{ .fw_name = "xtal", },
+			{ .hw = &sys.hw, },
+		},
+		.num_parents = 2,
+	},
+};
+
+static struct clk_regmap saradc_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = SAR_ADC_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "saradc_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&saradc_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap saradc = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SAR_ADC_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "saradc",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&saradc_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static const struct clk_parent_data pwm_abcd_parents[] = {
+	{ .fw_name = "xtal", },
+	{ .hw = &sys.hw },
+	{ .hw = &rtc.hw },
+};
+
+static struct clk_regmap pwm_a_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PWM_CLK_AB_CTRL,
+		.mask = 0x1,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_a_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = pwm_abcd_parents,
+		.num_parents = ARRAY_SIZE(pwm_abcd_parents),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap pwm_a_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PWM_CLK_AB_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_a_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_a_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_a = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PWM_CLK_AB_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "pwm_a",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_a_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_b_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PWM_CLK_AB_CTRL,
+		.mask = 0x1,
+		.shift = 25,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_b_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = pwm_abcd_parents,
+		.num_parents = ARRAY_SIZE(pwm_abcd_parents),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap pwm_b_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PWM_CLK_AB_CTRL,
+		.shift = 16,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_b_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_b_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_b = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PWM_CLK_AB_CTRL,
+		.bit_idx = 24,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "pwm_b",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_b_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_c_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PWM_CLK_CD_CTRL,
+		.mask = 0x1,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_c_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = pwm_abcd_parents,
+		.num_parents = ARRAY_SIZE(pwm_abcd_parents),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap pwm_c_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PWM_CLK_CD_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_c_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_c_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_c = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PWM_CLK_CD_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "pwm_c",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_c_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_d_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PWM_CLK_CD_CTRL,
+		.mask = 0x1,
+		.shift = 25,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_d_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = pwm_abcd_parents,
+		.num_parents = ARRAY_SIZE(pwm_abcd_parents),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap pwm_d_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PWM_CLK_CD_CTRL,
+		.shift = 16,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_d_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_d_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_d = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PWM_CLK_CD_CTRL,
+		.bit_idx = 24,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "pwm_d",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_d_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static const struct clk_parent_data pwm_ef_parents[] = {
+	{ .fw_name = "xtal", },
+	{ .hw = &sys.hw },
+	{ .fw_name = "fclk_div5", },
+	{ .hw = &rtc.hw },
+};
+
+static struct clk_regmap pwm_e_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PWM_CLK_EF_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_e_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = pwm_ef_parents,
+		.num_parents = ARRAY_SIZE(pwm_ef_parents),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap pwm_e_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PWM_CLK_EF_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_e_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_e_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_e = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PWM_CLK_EF_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "pwm_e",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_e_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_f_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PWM_CLK_EF_CTRL,
+		.mask = 0x3,
+		.shift = 25,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_f_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = pwm_ef_parents,
+		.num_parents = ARRAY_SIZE(pwm_ef_parents),
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap pwm_f_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PWM_CLK_EF_CTRL,
+		.shift = 16,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "pwm_f_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_f_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap pwm_f = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PWM_CLK_EF_CTRL,
+		.bit_idx = 24,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "pwm_f",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&pwm_f_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+/*
+ * spicc clk
+ *   fdiv2   |\         |\       _____
+ *  ---------| |---DIV--| |     |     |    spicc out
+ *  ---------| |        | |-----|GATE |---------
+ *     ..... |/         | /     |_____|
+ *  --------------------|/
+ *                 24M
+ */
+static const struct clk_parent_data spicc_spifc_parents[] = {
+	{ .fw_name = "fclk_div2"},
+	{ .fw_name = "fclk_div3"},
+	{ .fw_name = "fclk_div5"},
+	{ .fw_name = "hifi_pll" },
+};
+
+static struct clk_regmap spicc_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SPICC_CLK_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "spicc_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = spicc_spifc_parents,
+		.num_parents = ARRAY_SIZE(spicc_spifc_parents),
+	},
+};
+
+static struct clk_regmap spicc_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = SPICC_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "spicc_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&spicc_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap spicc_sel2 = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SPICC_CLK_CTRL,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "spicc_sel2",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = (const struct clk_parent_data []) {
+			{ .hw = &spicc_div.hw },
+			{ .fw_name = "xtal", },
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap spicc = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SPICC_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "spicc",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&spicc_sel2.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap ts_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = TS_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "ts_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap ts = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = TS_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "ts",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&ts_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap spifc_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SPIFC_CLK_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "spifc_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = spicc_spifc_parents,
+		.num_parents = ARRAY_SIZE(spicc_spifc_parents),
+	},
+};
+
+static struct clk_regmap spifc_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = SPIFC_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "spifc_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&spifc_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap spifc_sel2 = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SPIFC_CLK_CTRL,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "spifc_sel2",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = (const struct clk_parent_data []) {
+			{ .hw = &spifc_div.hw },
+			{ .fw_name = "xtal", },
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap spifc = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SPIFC_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "spifc",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&spifc_sel2.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static const struct clk_parent_data usb_bus_parents[] = {
+	{ .fw_name = "xtal", },
+	{ .hw = &sys.hw },
+	{ .fw_name = "fclk_div3", },
+	{ .fw_name = "fclk_div5", },
+};
+
+static struct clk_regmap usb_bus_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = USB_BUSCLK_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "usb_bus_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = usb_bus_parents,
+		.num_parents = ARRAY_SIZE(usb_bus_parents),
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap usb_bus_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = USB_BUSCLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "usb_bus_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&usb_bus_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap usb_bus = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = USB_BUSCLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "usb_bus",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&usb_bus_div.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static const struct clk_parent_data sd_emmc_psram_dmc_parents[] = {
+	{ .fw_name = "fclk_div2", },
+	{ .fw_name = "fclk_div3", },
+	{ .fw_name = "fclk_div5", },
+	{ .fw_name = "hifi_pll", },
+};
+
+static struct clk_regmap sd_emmc_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SD_EMMC_CLK_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sd_emmc_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = sd_emmc_psram_dmc_parents,
+		.num_parents = ARRAY_SIZE(sd_emmc_psram_dmc_parents),
+	},
+};
+
+static struct clk_regmap sd_emmc_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = SD_EMMC_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sd_emmc_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sd_emmc_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap sd_emmc_sel2 = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = SD_EMMC_CLK_CTRL,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "sd_emmc_sel2",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = (const struct clk_parent_data []) {
+			{ .hw = &sd_emmc_div.hw },
+			{ .fw_name = "xtal", },
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap sd_emmc = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = SD_EMMC_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "sd_emmc",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&sd_emmc_sel2.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap psram_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PSRAM_CLK_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "psram_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = sd_emmc_psram_dmc_parents,
+		.num_parents = ARRAY_SIZE(sd_emmc_psram_dmc_parents),
+	},
+};
+
+static struct clk_regmap psram_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = PSRAM_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "psram_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&psram_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap psram_sel2 = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = PSRAM_CLK_CTRL,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "psram_sel2",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = (const struct clk_parent_data []) {
+			{ .hw = &psram_div.hw },
+			{ .fw_name = "xtal", },
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap psram = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = PSRAM_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "psram",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&psram_sel2.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dmc_sel = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DMC_CLK_CTRL,
+		.mask = 0x3,
+		.shift = 9,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dmc_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = sd_emmc_psram_dmc_parents,
+		.num_parents = ARRAY_SIZE(sd_emmc_psram_dmc_parents),
+	},
+};
+
+static struct clk_regmap dmc_div = {
+	.data = &(struct clk_regmap_div_data){
+		.offset = DMC_CLK_CTRL,
+		.shift = 0,
+		.width = 8,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dmc_div",
+		.ops = &clk_regmap_divider_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dmc_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dmc_sel2 = {
+	.data = &(struct clk_regmap_mux_data){
+		.offset = DMC_CLK_CTRL,
+		.mask = 0x1,
+		.shift = 15,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "dmc_sel2",
+		.ops = &clk_regmap_mux_ops,
+		.parent_data = (const struct clk_parent_data []) {
+			{ .hw = &dmc_div.hw },
+			{ .fw_name = "xtal", },
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap dmc = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = DMC_CLK_CTRL,
+		.bit_idx = 8,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "dmc",
+		.ops = &clk_regmap_gate_ro_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&dmc_sel2.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap ceca_32k_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CECA_CLK_CTRL0,
+		.bit_idx = 31,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "ceca_32k_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap ceca_32k_div = {
+	.data = &(struct meson_clk_dualdiv_data){
+		.n1 = {
+			.reg_off = CECA_CLK_CTRL0,
+			.shift   = 0,
+			.width   = 12,
+		},
+		.n2 = {
+			.reg_off = CECA_CLK_CTRL0,
+			.shift   = 12,
+			.width   = 12,
+		},
+		.m1 = {
+			.reg_off = CECA_CLK_CTRL1,
+			.shift   = 0,
+			.width   = 12,
+		},
+		.m2 = {
+			.reg_off = CECA_CLK_CTRL1,
+			.shift   = 12,
+			.width   = 12,
+		},
+		.dual = {
+			.reg_off = CECA_CLK_CTRL0,
+			.shift   = 28,
+			.width   = 1,
+		},
+		.table = clk_32k_div_table,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "ceca_32k_div",
+		.ops = &meson_clk_dualdiv_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&ceca_32k_in.hw
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap ceca_32k_sel_pre = {
+	.data = &(struct clk_regmap_mux_data) {
+		.offset = CECA_CLK_CTRL1,
+		.mask = 0x1,
+		.shift = 24,
+		.flags = CLK_MUX_ROUND_CLOSEST,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "ceca_32k_sel_pre",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&ceca_32k_div.hw,
+			&ceca_32k_in.hw,
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap ceca_32k_sel = {
+	.data = &(struct clk_regmap_mux_data) {
+		.offset = CECA_CLK_CTRL1,
+		.mask = 0x1,
+		.shift = 31,
+		.flags = CLK_MUX_ROUND_CLOSEST,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "ceca_32k_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&ceca_32k_sel_pre.hw,
+			&rtc.hw,
+		},
+		.num_parents = 2,
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap ceca_32k_out = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CECA_CLK_CTRL0,
+		.bit_idx = 30,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "ceca_32k_out",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&ceca_32k_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap cecb_32k_in = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CECB_CLK_CTRL0,
+		.bit_idx = 31,
+	},
+	.hw.init = &(struct clk_init_data) {
+		.name = "cecb_32k_in",
+		.ops = &clk_regmap_gate_ops,
+		.parent_data = &(const struct clk_parent_data) {
+			.fw_name = "xtal",
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap cecb_32k_div = {
+	.data = &(struct meson_clk_dualdiv_data){
+		.n1 = {
+			.reg_off = CECB_CLK_CTRL0,
+			.shift   = 0,
+			.width   = 12,
+		},
+		.n2 = {
+			.reg_off = CECB_CLK_CTRL0,
+			.shift   = 12,
+			.width   = 12,
+		},
+		.m1 = {
+			.reg_off = CECB_CLK_CTRL1,
+			.shift   = 0,
+			.width   = 12,
+		},
+		.m2 = {
+			.reg_off = CECB_CLK_CTRL1,
+			.shift   = 12,
+			.width   = 12,
+		},
+		.dual = {
+			.reg_off = CECB_CLK_CTRL0,
+			.shift   = 28,
+			.width   = 1,
+		},
+		.table = clk_32k_div_table,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "cecb_32k_div",
+		.ops = &meson_clk_dualdiv_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&cecb_32k_in.hw
+		},
+		.num_parents = 1,
+	},
+};
+
+static struct clk_regmap cecb_32k_sel_pre = {
+	.data = &(struct clk_regmap_mux_data) {
+		.offset = CECB_CLK_CTRL1,
+		.mask = 0x1,
+		.shift = 24,
+		.flags = CLK_MUX_ROUND_CLOSEST,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "cecb_32k_sel_pre",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&cecb_32k_div.hw,
+			&cecb_32k_in.hw,
+		},
+		.num_parents = 2,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+static struct clk_regmap cecb_32k_sel = {
+	.data = &(struct clk_regmap_mux_data) {
+		.offset = CECB_CLK_CTRL1,
+		.mask = 0x1,
+		.shift = 31,
+		.flags = CLK_MUX_ROUND_CLOSEST,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "cecb_32k_sel",
+		.ops = &clk_regmap_mux_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&cecb_32k_sel_pre.hw,
+			&rtc.hw,
+		},
+		.num_parents = 2,
+		/* For more information, please refer to rtc clock */
+		.flags = CLK_SET_RATE_NO_REPARENT,
+	},
+};
+
+static struct clk_regmap cecb_32k_out = {
+	.data = &(struct clk_regmap_gate_data){
+		.offset = CECB_CLK_CTRL0,
+		.bit_idx = 30,
+	},
+	.hw.init = &(struct clk_init_data){
+		.name = "cecb_32k_out",
+		.ops = &clk_regmap_gate_ops,
+		.parent_hws = (const struct clk_hw *[]) {
+			&cecb_32k_sel.hw
+		},
+		.num_parents = 1,
+		.flags = CLK_SET_RATE_PARENT,
+	},
+};
+
+#define MESON_GATE(_name, _reg, _bit) \
+	MESON_PCLK(_name, _reg, _bit, &sys.hw)
+
+static MESON_GATE(clktree,	SYS_CLK_EN0,	0);
+static MESON_GATE(reset_ctrl,	SYS_CLK_EN0,	1);
+static MESON_GATE(analog_ctrl,	SYS_CLK_EN0,	2);
+static MESON_GATE(pwr_ctrl,	SYS_CLK_EN0,	3);
+static MESON_GATE(pad_ctrl,	SYS_CLK_EN0,	4);
+static MESON_GATE(sys_ctrl,	SYS_CLK_EN0,	5);
+static MESON_GATE(temp_sensor,	SYS_CLK_EN0,	6);
+static MESON_GATE(am2axi_dev,	SYS_CLK_EN0,	7);
+static MESON_GATE(spicc_b,	SYS_CLK_EN0,	8);
+static MESON_GATE(spicc_a,	SYS_CLK_EN0,	9);
+static MESON_GATE(msr,		SYS_CLK_EN0,	10);
+static MESON_GATE(audio,	SYS_CLK_EN0,	11);
+static MESON_GATE(jtag_ctrl,	SYS_CLK_EN0,	12);
+static MESON_GATE(saradc_en,	SYS_CLK_EN0,	13);
+static MESON_GATE(pwm_ef,	SYS_CLK_EN0,	14);
+static MESON_GATE(pwm_cd,	SYS_CLK_EN0,	15);
+static MESON_GATE(pwm_ab,	SYS_CLK_EN0,	16);
+static MESON_GATE(cec,		SYS_CLK_EN0,	17);
+static MESON_GATE(i2c_s,	SYS_CLK_EN0,	18);
+static MESON_GATE(ir_ctrl,	SYS_CLK_EN0,	19);
+static MESON_GATE(i2c_m_d,	SYS_CLK_EN0,	20);
+static MESON_GATE(i2c_m_c,	SYS_CLK_EN0,	21);
+static MESON_GATE(i2c_m_b,	SYS_CLK_EN0,	22);
+static MESON_GATE(i2c_m_a,	SYS_CLK_EN0,	23);
+static MESON_GATE(acodec,	SYS_CLK_EN0,	24);
+static MESON_GATE(otp,		SYS_CLK_EN0,	25);
+static MESON_GATE(sd_emmc_a,	SYS_CLK_EN0,	26);
+static MESON_GATE(usb_phy,	SYS_CLK_EN0,	27);
+static MESON_GATE(usb_ctrl,	SYS_CLK_EN0,	28);
+static MESON_GATE(sys_dspb,	SYS_CLK_EN0,	29);
+static MESON_GATE(sys_dspa,	SYS_CLK_EN0,	30);
+static MESON_GATE(dma,		SYS_CLK_EN0,	31);
+static MESON_GATE(irq_ctrl,	SYS_CLK_EN1,	0);
+static MESON_GATE(nic,		SYS_CLK_EN1,	1);
+static MESON_GATE(gic,		SYS_CLK_EN1,	2);
+static MESON_GATE(uart_c,	SYS_CLK_EN1,	3);
+static MESON_GATE(uart_b,	SYS_CLK_EN1,	4);
+static MESON_GATE(uart_a,	SYS_CLK_EN1,	5);
+static MESON_GATE(sys_psram,	SYS_CLK_EN1,	6);
+static MESON_GATE(rsa,		SYS_CLK_EN1,	8);
+static MESON_GATE(coresight,	SYS_CLK_EN1,	9);
+static MESON_GATE(am2axi_vad,	AXI_CLK_EN,	0);
+static MESON_GATE(audio_vad,	AXI_CLK_EN,	1);
+static MESON_GATE(axi_dmc,	AXI_CLK_EN,	3);
+static MESON_GATE(axi_psram,	AXI_CLK_EN,	4);
+static MESON_GATE(ramb,		AXI_CLK_EN,	5);
+static MESON_GATE(rama,		AXI_CLK_EN,	6);
+static MESON_GATE(axi_spifc,	AXI_CLK_EN,	7);
+static MESON_GATE(axi_nic,	AXI_CLK_EN,	8);
+static MESON_GATE(axi_dma,	AXI_CLK_EN,	9);
+static MESON_GATE(cpu_ctrl,	AXI_CLK_EN,	10);
+static MESON_GATE(rom,		AXI_CLK_EN,	11);
+static MESON_GATE(prod_i2c,	AXI_CLK_EN,	12);
+
+/* Array of all clocks registered by this provider */
+static struct clk_hw_onecell_data a1_periphs_clks = {
+	.hws = {
+		[CLKID_XTAL_IN]			= &xtal_in.hw,
+		[CLKID_FIXPLL_IN]		= &fixpll_in.hw,
+		[CLKID_USB_PHY_IN]		= &usb_phy_in.hw,
+		[CLKID_USB_CTRL_IN]		= &usb_ctrl_in.hw,
+		[CLKID_HIFIPLL_IN]		= &hifipll_in.hw,
+		[CLKID_SYSPLL_IN]		= &syspll_in.hw,
+		[CLKID_DDS_IN]			= &dds_in.hw,
+		[CLKID_SYS]			= &sys.hw,
+		[CLKID_CLKTREE]			= &clktree.hw,
+		[CLKID_RESET_CTRL]		= &reset_ctrl.hw,
+		[CLKID_ANALOG_CTRL]		= &analog_ctrl.hw,
+		[CLKID_PWR_CTRL]		= &pwr_ctrl.hw,
+		[CLKID_PAD_CTRL]		= &pad_ctrl.hw,
+		[CLKID_SYS_CTRL]		= &sys_ctrl.hw,
+		[CLKID_TEMP_SENSOR]		= &temp_sensor.hw,
+		[CLKID_AM2AXI_DIV]		= &am2axi_dev.hw,
+		[CLKID_SPICC_B]			= &spicc_b.hw,
+		[CLKID_SPICC_A]			= &spicc_a.hw,
+		[CLKID_MSR]			= &msr.hw,
+		[CLKID_AUDIO]			= &audio.hw,
+		[CLKID_JTAG_CTRL]		= &jtag_ctrl.hw,
+		[CLKID_SARADC_EN]		= &saradc_en.hw,
+		[CLKID_PWM_EF]			= &pwm_ef.hw,
+		[CLKID_PWM_CD]			= &pwm_cd.hw,
+		[CLKID_PWM_AB]			= &pwm_ab.hw,
+		[CLKID_CEC]			= &cec.hw,
+		[CLKID_I2C_S]			= &i2c_s.hw,
+		[CLKID_IR_CTRL]			= &ir_ctrl.hw,
+		[CLKID_I2C_M_D]			= &i2c_m_d.hw,
+		[CLKID_I2C_M_C]			= &i2c_m_c.hw,
+		[CLKID_I2C_M_B]			= &i2c_m_b.hw,
+		[CLKID_I2C_M_A]			= &i2c_m_a.hw,
+		[CLKID_ACODEC]			= &acodec.hw,
+		[CLKID_OTP]			= &otp.hw,
+		[CLKID_SD_EMMC_A]		= &sd_emmc_a.hw,
+		[CLKID_USB_PHY]			= &usb_phy.hw,
+		[CLKID_USB_CTRL]		= &usb_ctrl.hw,
+		[CLKID_SYS_DSPB]		= &sys_dspb.hw,
+		[CLKID_SYS_DSPA]		= &sys_dspa.hw,
+		[CLKID_DMA]			= &dma.hw,
+		[CLKID_IRQ_CTRL]		= &irq_ctrl.hw,
+		[CLKID_NIC]			= &nic.hw,
+		[CLKID_GIC]			= &gic.hw,
+		[CLKID_UART_C]			= &uart_c.hw,
+		[CLKID_UART_B]			= &uart_b.hw,
+		[CLKID_UART_A]			= &uart_a.hw,
+		[CLKID_SYS_PSRAM]		= &sys_psram.hw,
+		[CLKID_RSA]			= &rsa.hw,
+		[CLKID_CORESIGHT]		= &coresight.hw,
+		[CLKID_AM2AXI_VAD]		= &am2axi_vad.hw,
+		[CLKID_AUDIO_VAD]		= &audio_vad.hw,
+		[CLKID_AXI_DMC]			= &axi_dmc.hw,
+		[CLKID_AXI_PSRAM]		= &axi_psram.hw,
+		[CLKID_RAMB]			= &ramb.hw,
+		[CLKID_RAMA]			= &rama.hw,
+		[CLKID_AXI_SPIFC]		= &axi_spifc.hw,
+		[CLKID_AXI_NIC]			= &axi_nic.hw,
+		[CLKID_AXI_DMA]			= &axi_dma.hw,
+		[CLKID_CPU_CTRL]		= &cpu_ctrl.hw,
+		[CLKID_ROM]			= &rom.hw,
+		[CLKID_PROC_I2C]		= &prod_i2c.hw,
+		[CLKID_DSPA_SEL]		= &dspa_sel.hw,
+		[CLKID_DSPB_SEL]		= &dspb_sel.hw,
+		[CLKID_DSPA_EN]			= &dspa_en.hw,
+		[CLKID_DSPA_EN_NIC]		= &dspa_en_nic.hw,
+		[CLKID_DSPB_EN]			= &dspb_en.hw,
+		[CLKID_DSPB_EN_NIC]		= &dspb_en_nic.hw,
+		[CLKID_RTC]			= &rtc.hw,
+		[CLKID_CECA_32K]		= &ceca_32k_out.hw,
+		[CLKID_CECB_32K]		= &cecb_32k_out.hw,
+		[CLKID_24M]			= &clk_24m.hw,
+		[CLKID_12M]			= &clk_12m.hw,
+		[CLKID_FCLK_DIV2_DIVN]		= &fclk_div2_divn.hw,
+		[CLKID_GEN]			= &gen.hw,
+		[CLKID_SARADC_SEL]		= &saradc_sel.hw,
+		[CLKID_SARADC]			= &saradc.hw,
+		[CLKID_PWM_A]			= &pwm_a.hw,
+		[CLKID_PWM_B]			= &pwm_b.hw,
+		[CLKID_PWM_C]			= &pwm_c.hw,
+		[CLKID_PWM_D]			= &pwm_d.hw,
+		[CLKID_PWM_E]			= &pwm_e.hw,
+		[CLKID_PWM_F]			= &pwm_f.hw,
+		[CLKID_SPICC]			= &spicc.hw,
+		[CLKID_TS]			= &ts.hw,
+		[CLKID_SPIFC]			= &spifc.hw,
+		[CLKID_USB_BUS]			= &usb_bus.hw,
+		[CLKID_SD_EMMC]			= &sd_emmc.hw,
+		[CLKID_PSRAM]			= &psram.hw,
+		[CLKID_DMC]			= &dmc.hw,
+		[CLKID_SYS_A_SEL]		= &sys_a_sel.hw,
+		[CLKID_SYS_A_DIV]		= &sys_a_div.hw,
+		[CLKID_SYS_A]			= &sys_a.hw,
+		[CLKID_SYS_B_SEL]		= &sys_b_sel.hw,
+		[CLKID_SYS_B_DIV]		= &sys_b_div.hw,
+		[CLKID_SYS_B]			= &sys_b.hw,
+		[CLKID_DSPA_A_SEL]		= &dspa_a_sel.hw,
+		[CLKID_DSPA_A_DIV]		= &dspa_a_div.hw,
+		[CLKID_DSPA_A]			= &dspa_a.hw,
+		[CLKID_DSPA_B_SEL]		= &dspa_b_sel.hw,
+		[CLKID_DSPA_B_DIV]		= &dspa_b_div.hw,
+		[CLKID_DSPA_B]			= &dspa_b.hw,
+		[CLKID_DSPB_A_SEL]		= &dspb_a_sel.hw,
+		[CLKID_DSPB_A_DIV]		= &dspb_a_div.hw,
+		[CLKID_DSPB_A]			= &dspb_a.hw,
+		[CLKID_DSPB_B_SEL]		= &dspb_b_sel.hw,
+		[CLKID_DSPB_B_DIV]		= &dspb_b_div.hw,
+		[CLKID_DSPB_B]			= &dspb_b.hw,
+		[CLKID_RTC_32K_IN]		= &rtc_32k_in.hw,
+		[CLKID_RTC_32K_DIV]		= &rtc_32k_div.hw,
+		[CLKID_RTC_32K_XTAL]		= &rtc_32k_xtal.hw,
+		[CLKID_RTC_32K_SEL]		= &rtc_32k_sel.hw,
+		[CLKID_CECB_32K_IN]		= &cecb_32k_in.hw,
+		[CLKID_CECB_32K_DIV]		= &cecb_32k_div.hw,
+		[CLKID_CECB_32K_SEL_PRE]	= &cecb_32k_sel_pre.hw,
+		[CLKID_CECB_32K_SEL]		= &cecb_32k_sel.hw,
+		[CLKID_CECA_32K_IN]		= &ceca_32k_in.hw,
+		[CLKID_CECA_32K_DIV]		= &ceca_32k_div.hw,
+		[CLKID_CECA_32K_SEL_PRE]	= &ceca_32k_sel_pre.hw,
+		[CLKID_CECA_32K_SEL]		= &ceca_32k_sel.hw,
+		[CLKID_DIV2_PRE]		= &fclk_div2_divn_pre.hw,
+		[CLKID_24M_DIV2]		= &clk_24m_div2.hw,
+		[CLKID_GEN_SEL]			= &gen_sel.hw,
+		[CLKID_GEN_DIV]			= &gen_div.hw,
+		[CLKID_SARADC_DIV]		= &saradc_div.hw,
+		[CLKID_PWM_A_SEL]		= &pwm_a_sel.hw,
+		[CLKID_PWM_A_DIV]		= &pwm_a_div.hw,
+		[CLKID_PWM_B_SEL]		= &pwm_b_sel.hw,
+		[CLKID_PWM_B_DIV]		= &pwm_b_div.hw,
+		[CLKID_PWM_C_SEL]		= &pwm_c_sel.hw,
+		[CLKID_PWM_C_DIV]		= &pwm_c_div.hw,
+		[CLKID_PWM_D_SEL]		= &pwm_d_sel.hw,
+		[CLKID_PWM_D_DIV]		= &pwm_d_div.hw,
+		[CLKID_PWM_E_SEL]		= &pwm_e_sel.hw,
+		[CLKID_PWM_E_DIV]		= &pwm_e_div.hw,
+		[CLKID_PWM_F_SEL]		= &pwm_f_sel.hw,
+		[CLKID_PWM_F_DIV]		= &pwm_f_div.hw,
+		[CLKID_SPICC_SEL]		= &spicc_sel.hw,
+		[CLKID_SPICC_DIV]		= &spicc_div.hw,
+		[CLKID_SPICC_SEL2]		= &spicc_sel2.hw,
+		[CLKID_TS_DIV]			= &ts_div.hw,
+		[CLKID_SPIFC_SEL]		= &spifc_sel.hw,
+		[CLKID_SPIFC_DIV]		= &spifc_div.hw,
+		[CLKID_SPIFC_SEL2]		= &spifc_sel2.hw,
+		[CLKID_USB_BUS_SEL]		= &usb_bus_sel.hw,
+		[CLKID_USB_BUS_DIV]		= &usb_bus_div.hw,
+		[CLKID_SD_EMMC_SEL]		= &sd_emmc_sel.hw,
+		[CLKID_SD_EMMC_DIV]		= &sd_emmc_div.hw,
+		[CLKID_SD_EMMC_SEL2]		= &sd_emmc_sel2.hw,
+		[CLKID_PSRAM_SEL]		= &psram_sel.hw,
+		[CLKID_PSRAM_DIV]		= &psram_div.hw,
+		[CLKID_PSRAM_SEL2]		= &psram_sel2.hw,
+		[CLKID_DMC_SEL]			= &dmc_sel.hw,
+		[CLKID_DMC_DIV]			= &dmc_div.hw,
+		[CLKID_DMC_SEL2]		= &dmc_sel2.hw,
+		[NR_CLKS]			= NULL,
+	},
+	.num = NR_CLKS,
+};
+
+/* Convenience table to populate regmap in .probe */
+static struct clk_regmap *const a1_periphs_regmaps[] = {
+	&xtal_in,
+	&fixpll_in,
+	&usb_phy_in,
+	&usb_ctrl_in,
+	&hifipll_in,
+	&syspll_in,
+	&dds_in,
+	&sys,
+	&clktree,
+	&reset_ctrl,
+	&analog_ctrl,
+	&pwr_ctrl,
+	&pad_ctrl,
+	&sys_ctrl,
+	&temp_sensor,
+	&am2axi_dev,
+	&spicc_b,
+	&spicc_a,
+	&msr,
+	&audio,
+	&jtag_ctrl,
+	&saradc_en,
+	&pwm_ef,
+	&pwm_cd,
+	&pwm_ab,
+	&cec,
+	&i2c_s,
+	&ir_ctrl,
+	&i2c_m_d,
+	&i2c_m_c,
+	&i2c_m_b,
+	&i2c_m_a,
+	&acodec,
+	&otp,
+	&sd_emmc_a,
+	&usb_phy,
+	&usb_ctrl,
+	&sys_dspb,
+	&sys_dspa,
+	&dma,
+	&irq_ctrl,
+	&nic,
+	&gic,
+	&uart_c,
+	&uart_b,
+	&uart_a,
+	&sys_psram,
+	&rsa,
+	&coresight,
+	&am2axi_vad,
+	&audio_vad,
+	&axi_dmc,
+	&axi_psram,
+	&ramb,
+	&rama,
+	&axi_spifc,
+	&axi_nic,
+	&axi_dma,
+	&cpu_ctrl,
+	&rom,
+	&prod_i2c,
+	&dspa_sel,
+	&dspb_sel,
+	&dspa_en,
+	&dspa_en_nic,
+	&dspb_en,
+	&dspb_en_nic,
+	&rtc,
+	&ceca_32k_out,
+	&cecb_32k_out,
+	&clk_24m,
+	&clk_12m,
+	&fclk_div2_divn,
+	&gen,
+	&saradc_sel,
+	&saradc,
+	&pwm_a,
+	&pwm_b,
+	&pwm_c,
+	&pwm_d,
+	&pwm_e,
+	&pwm_f,
+	&spicc,
+	&ts,
+	&spifc,
+	&usb_bus,
+	&sd_emmc,
+	&psram,
+	&dmc,
+	&sys_a_sel,
+	&sys_a_div,
+	&sys_a,
+	&sys_b_sel,
+	&sys_b_div,
+	&sys_b,
+	&dspa_a_sel,
+	&dspa_a_div,
+	&dspa_a,
+	&dspa_b_sel,
+	&dspa_b_div,
+	&dspa_b,
+	&dspb_a_sel,
+	&dspb_a_div,
+	&dspb_a,
+	&dspb_b_sel,
+	&dspb_b_div,
+	&dspb_b,
+	&rtc_32k_in,
+	&rtc_32k_div,
+	&rtc_32k_xtal,
+	&rtc_32k_sel,
+	&cecb_32k_in,
+	&cecb_32k_div,
+	&cecb_32k_sel_pre,
+	&cecb_32k_sel,
+	&ceca_32k_in,
+	&ceca_32k_div,
+	&ceca_32k_sel_pre,
+	&ceca_32k_sel,
+	&fclk_div2_divn_pre,
+	&gen_sel,
+	&gen_div,
+	&saradc_div,
+	&pwm_a_sel,
+	&pwm_a_div,
+	&pwm_b_sel,
+	&pwm_b_div,
+	&pwm_c_sel,
+	&pwm_c_div,
+	&pwm_d_sel,
+	&pwm_d_div,
+	&pwm_e_sel,
+	&pwm_e_div,
+	&pwm_f_sel,
+	&pwm_f_div,
+	&spicc_sel,
+	&spicc_div,
+	&spicc_sel2,
+	&ts_div,
+	&spifc_sel,
+	&spifc_div,
+	&spifc_sel2,
+	&usb_bus_sel,
+	&usb_bus_div,
+	&sd_emmc_sel,
+	&sd_emmc_div,
+	&sd_emmc_sel2,
+	&psram_sel,
+	&psram_div,
+	&psram_sel2,
+	&dmc_sel,
+	&dmc_div,
+	&dmc_sel2,
+};
+
+static struct regmap_config a1_periphs_regmap_cfg = {
+	.reg_bits   = 32,
+	.val_bits   = 32,
+	.reg_stride = 4,
+};
+
+static int meson_a1_periphs_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	void __iomem *base;
+	struct regmap *map;
+	int clkid, i, err;
+
+	base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(base))
+		return dev_err_probe(dev, PTR_ERR(base),
+				     "can't ioremap resource\n");
+
+	map = devm_regmap_init_mmio(dev, base, &a1_periphs_regmap_cfg);
+	if (IS_ERR(map))
+		return dev_err_probe(dev, PTR_ERR(map),
+				     "can't init regmap mmio region\n");
+
+	/* Populate regmap for the regmap backed clocks */
+	for (i = 0; i < ARRAY_SIZE(a1_periphs_regmaps); i++)
+		a1_periphs_regmaps[i]->map = map;
+
+	for (clkid = 0; clkid < a1_periphs_clks.num; clkid++) {
+		err = devm_clk_hw_register(dev, a1_periphs_clks.hws[clkid]);
+		if (err)
+			return dev_err_probe(dev, err,
+					     "clock[%d] registration failed\n",
+					     clkid);
+	}
+
+	return devm_of_clk_add_hw_provider(dev, of_clk_hw_onecell_get,
+					   &a1_periphs_clks);
+}
+
+static const struct of_device_id a1_periphs_clkc_match_table[] = {
+	{ .compatible = "amlogic,a1-clkc", },
+	{}
+};
+MODULE_DEVICE_TABLE(of, a1_periphs_clkc_match_table);
+
+static struct platform_driver a1_periphs_clkc_driver = {
+	.probe = meson_a1_periphs_probe,
+	.driver = {
+		.name = "a1-clkc",
+		.of_match_table = a1_periphs_clkc_match_table,
+	},
+};
+
+module_platform_driver(a1_periphs_clkc_driver);
+MODULE_AUTHOR("Jian Hu <[e-mail]>");
+MODULE_AUTHOR("Dmitry Rokosov <[e-mail]>");
+MODULE_LICENSE("GPL");
diff --git a/drivers/clk/meson/a1.h b/drivers/clk/meson/a1.h
new file mode 100644
index 000000000000..359b46b73035
--- /dev/null
+++ b/drivers/clk/meson/a1.h
@@ -0,0 +1,114 @@ 
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ * Amlogic A1 Peripheral Clock Controller internals
+ *
+ * Copyright (c) 2019 Amlogic, Inc. All rights reserved.
+ * Author: Jian Hu <[e-mail]>
+ *
+ * Copyright (c) 2023, SberDevices. All Rights Reserved.
+ * Author: Dmitry Rokosov <[e-mail]>
+ */
+
+#ifndef __A1_H
+#define __A1_H
+
+/* peripheral clock controller register offset */
+#define SYS_OSCIN_CTRL		0x0
+#define RTC_BY_OSCIN_CTRL0	0x4
+#define RTC_BY_OSCIN_CTRL1	0x8
+#define RTC_CTRL		0xc
+#define SYS_CLK_CTRL0		0x10
+#define SYS_CLK_EN0		0x1c
+#define SYS_CLK_EN1		0x20
+#define AXI_CLK_EN		0x24
+#define DSPA_CLK_EN		0x28
+#define DSPB_CLK_EN		0x2c
+#define DSPA_CLK_CTRL0		0x30
+#define DSPB_CLK_CTRL0		0x34
+#define CLK12_24_CTRL		0x38
+#define GEN_CLK_CTRL		0x3c
+#define SAR_ADC_CLK_CTRL	0xc0
+#define PWM_CLK_AB_CTRL		0xc4
+#define PWM_CLK_CD_CTRL		0xc8
+#define PWM_CLK_EF_CTRL		0xcc
+#define SPICC_CLK_CTRL		0xd0
+#define TS_CLK_CTRL		0xd4
+#define SPIFC_CLK_CTRL		0xd8
+#define USB_BUSCLK_CTRL		0xdc
+#define SD_EMMC_CLK_CTRL	0xe0
+#define CECA_CLK_CTRL0		0xe4
+#define CECA_CLK_CTRL1		0xe8
+#define CECB_CLK_CTRL0		0xec
+#define CECB_CLK_CTRL1		0xf0
+#define PSRAM_CLK_CTRL		0xf4
+#define DMC_CLK_CTRL		0xf8
+
+/* include the CLKIDs that have been made part of the DT binding */
+#include <dt-bindings/clock/amlogic,a1-clkc.h>
+
+/*
+ * CLKID index values for internal clocks
+ *
+ * These indices are entirely contrived and do not map onto the hardware.
+ * It has now been decided to expose everything by default in the DT header:
+ * include/dt-bindings/clock/a1-clkc.h. Only the clocks ids we don't want
+ * to expose, such as the internal muxes and dividers of composite clocks,
+ * will remain defined here.
+ */
+#define CLKID_XTAL_IN		0
+#define CLKID_DSPA_SEL		61
+#define CLKID_DSPB_SEL		62
+#define CLKID_SARADC_SEL	74
+#define CLKID_SYS_A_SEL		89
+#define CLKID_SYS_A_DIV		90
+#define CLKID_SYS_A		91
+#define CLKID_SYS_B_SEL		92
+#define CLKID_SYS_B_DIV		93
+#define CLKID_SYS_B		94
+#define CLKID_DSPA_A_DIV	96
+#define CLKID_DSPA_A		97
+#define CLKID_DSPA_B_DIV	99
+#define CLKID_DSPA_B		100
+#define CLKID_DSPB_A_DIV	102
+#define CLKID_DSPB_A		103
+#define CLKID_DSPB_B_DIV	105
+#define CLKID_DSPB_B		106
+#define CLKID_RTC_32K_IN	107
+#define CLKID_RTC_32K_DIV	108
+#define CLKID_RTC_32K_XTAL	109
+#define CLKID_RTC_32K_SEL	110
+#define CLKID_CECB_32K_IN	111
+#define CLKID_CECB_32K_DIV	112
+#define CLKID_CECA_32K_IN	115
+#define CLKID_CECA_32K_DIV	116
+#define CLKID_DIV2_PRE		119
+#define CLKID_24M_DIV2		120
+#define CLKID_GEN_DIV		122
+#define CLKID_SARADC_DIV	123
+#define CLKID_PWM_A_DIV		125
+#define CLKID_PWM_B_DIV		127
+#define CLKID_PWM_C_DIV		129
+#define CLKID_PWM_D_DIV		131
+#define CLKID_PWM_E_DIV		133
+#define CLKID_PWM_F_DIV		135
+#define CLKID_SPICC_SEL		136
+#define CLKID_SPICC_DIV		137
+#define CLKID_SPICC_SEL2	138
+#define CLKID_TS_DIV		139
+#define CLKID_SPIFC_SEL		140
+#define CLKID_SPIFC_DIV		141
+#define CLKID_SPIFC_SEL2	142
+#define CLKID_USB_BUS_SEL	143
+#define CLKID_USB_BUS_DIV	144
+#define CLKID_SD_EMMC_SEL	145
+#define CLKID_SD_EMMC_DIV	146
+#define CLKID_SD_EMMC_SEL2	147
+#define CLKID_PSRAM_SEL		148
+#define CLKID_PSRAM_DIV		149
+#define CLKID_PSRAM_SEL2	150
+#define CLKID_DMC_SEL		151
+#define CLKID_DMC_DIV		152
+#define CLKID_DMC_SEL2		153
+#define NR_CLKS			154
+
+#endif /* __A1_H */