clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
an rpi, simply fetch it once.
Signed-off-by: Sean Young <[email protected]>
---
drivers/pwm/pwm-bcm2835.c | 21 ++++++++++++---------
1 file changed, 12 insertions(+), 9 deletions(-)
diff --git a/drivers/pwm/pwm-bcm2835.c b/drivers/pwm/pwm-bcm2835.c
index bdfc2a5ec0d6..59ea154dd657 100644
--- a/drivers/pwm/pwm-bcm2835.c
+++ b/drivers/pwm/pwm-bcm2835.c
@@ -28,6 +28,7 @@ struct bcm2835_pwm {
struct device *dev;
void __iomem *base;
struct clk *clk;
+ unsigned long rate;
};
static inline struct bcm2835_pwm *to_bcm2835_pwm(struct pwm_chip *chip)
@@ -63,17 +64,11 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
{
struct bcm2835_pwm *pc = to_bcm2835_pwm(chip);
- unsigned long rate = clk_get_rate(pc->clk);
unsigned long long period_cycles;
u64 max_period;
u32 val;
- if (!rate) {
- dev_err(pc->dev, "failed to get clock rate\n");
- return -EINVAL;
- }
-
/*
* period_cycles must be a 32 bit value, so period * rate / NSEC_PER_SEC
* must be <= U32_MAX. As U32_MAX * NSEC_PER_SEC < U64_MAX the
@@ -88,13 +83,13 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
* <=> period < ((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate
* <=> period <= ceil((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate) - 1
*/
- max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC / 2, rate) - 1;
+ max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC / 2, pc->rate) - 1;
if (state->period > max_period)
return -EINVAL;
/* set period */
- period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * rate, NSEC_PER_SEC);
+ period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * pc->rate, NSEC_PER_SEC);
/* don't accept a period that is too small */
if (period_cycles < PERIOD_MIN)
@@ -103,7 +98,7 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
writel(period_cycles, pc->base + PERIOD(pwm->hwpwm));
/* set duty cycle */
- val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * rate, NSEC_PER_SEC);
+ val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * pc->rate, NSEC_PER_SEC);
writel(val, pc->base + DUTY(pwm->hwpwm));
/* set polarity */
@@ -129,6 +124,7 @@ static const struct pwm_ops bcm2835_pwm_ops = {
.request = bcm2835_pwm_request,
.free = bcm2835_pwm_free,
.apply = bcm2835_pwm_apply,
+ .atomic = true,
.owner = THIS_MODULE,
};
@@ -156,6 +152,13 @@ static int bcm2835_pwm_probe(struct platform_device *pdev)
if (ret)
return ret;
+ pc->rate = clk_get_rate(pc->clk);
+ if (!pc->rate) {
+ dev_err(pc->dev, "failed to get clock rate\n");
+ ret = -EINVAL;
+ goto add_fail;
+ }
+
pc->chip.dev = &pdev->dev;
pc->chip.ops = &bcm2835_pwm_ops;
pc->chip.npwm = 2;
--
2.42.0
Hi Sean,
Am 13.10.23 um 12:46 schrieb Sean Young:
> clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
> an rpi, simply fetch it once.
does it mean you checked all possible SoCs (BCM2835, BCM2836, BCM2837,
BCM2711, BCM2712) for this change?
Is it impossible that the real clock can never be influenced by turbo
mode like SPI?
Best regards
>
> Signed-off-by: Sean Young <[email protected]>
> ---
> drivers/pwm/pwm-bcm2835.c | 21 ++++++++++++---------
> 1 file changed, 12 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/pwm/pwm-bcm2835.c b/drivers/pwm/pwm-bcm2835.c
> index bdfc2a5ec0d6..59ea154dd657 100644
> --- a/drivers/pwm/pwm-bcm2835.c
> +++ b/drivers/pwm/pwm-bcm2835.c
> @@ -28,6 +28,7 @@ struct bcm2835_pwm {
> struct device *dev;
> void __iomem *base;
> struct clk *clk;
> + unsigned long rate;
> };
>
> static inline struct bcm2835_pwm *to_bcm2835_pwm(struct pwm_chip *chip)
> @@ -63,17 +64,11 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> {
>
> struct bcm2835_pwm *pc = to_bcm2835_pwm(chip);
> - unsigned long rate = clk_get_rate(pc->clk);
> unsigned long long period_cycles;
> u64 max_period;
>
> u32 val;
>
> - if (!rate) {
> - dev_err(pc->dev, "failed to get clock rate\n");
> - return -EINVAL;
> - }
> -
> /*
> * period_cycles must be a 32 bit value, so period * rate / NSEC_PER_SEC
> * must be <= U32_MAX. As U32_MAX * NSEC_PER_SEC < U64_MAX the
> @@ -88,13 +83,13 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> * <=> period < ((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate
> * <=> period <= ceil((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate) - 1
> */
> - max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC / 2, rate) - 1;
> + max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC / 2, pc->rate) - 1;
>
> if (state->period > max_period)
> return -EINVAL;
>
> /* set period */
> - period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * rate, NSEC_PER_SEC);
> + period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * pc->rate, NSEC_PER_SEC);
>
> /* don't accept a period that is too small */
> if (period_cycles < PERIOD_MIN)
> @@ -103,7 +98,7 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> writel(period_cycles, pc->base + PERIOD(pwm->hwpwm));
>
> /* set duty cycle */
> - val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * rate, NSEC_PER_SEC);
> + val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * pc->rate, NSEC_PER_SEC);
> writel(val, pc->base + DUTY(pwm->hwpwm));
>
> /* set polarity */
> @@ -129,6 +124,7 @@ static const struct pwm_ops bcm2835_pwm_ops = {
> .request = bcm2835_pwm_request,
> .free = bcm2835_pwm_free,
> .apply = bcm2835_pwm_apply,
> + .atomic = true,
> .owner = THIS_MODULE,
> };
>
> @@ -156,6 +152,13 @@ static int bcm2835_pwm_probe(struct platform_device *pdev)
> if (ret)
> return ret;
>
> + pc->rate = clk_get_rate(pc->clk);
> + if (!pc->rate) {
> + dev_err(pc->dev, "failed to get clock rate\n");
> + ret = -EINVAL;
> + goto add_fail;
> + }
> +
> pc->chip.dev = &pdev->dev;
> pc->chip.ops = &bcm2835_pwm_ops;
> pc->chip.npwm = 2;
Hi,
Am Freitag, 13. Oktober 2023, 13:04:48 CEST schrieb Stefan Wahren:
> Hi Sean,
>
> Am 13.10.23 um 12:46 schrieb Sean Young:
> > clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
> > an rpi, simply fetch it once.
>
> does it mean you checked all possible SoCs (BCM2835, BCM2836, BCM2837,
> BCM2711, BCM2712) for this change?
>
> Is it impossible that the real clock can never be influenced by turbo
> mode like SPI?
Assuming the clock can change, which I would, then a clock notifier seems
appropriate. See [1] for an example.
Best regards,
Alexander
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/
commit/?id=90ad2cbe88c22d0215225ab9594eeead0eb24fde
> Best regards
>
> > Signed-off-by: Sean Young <[email protected]>
> > ---
> >
> > drivers/pwm/pwm-bcm2835.c | 21 ++++++++++++---------
> > 1 file changed, 12 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/pwm/pwm-bcm2835.c b/drivers/pwm/pwm-bcm2835.c
> > index bdfc2a5ec0d6..59ea154dd657 100644
> > --- a/drivers/pwm/pwm-bcm2835.c
> > +++ b/drivers/pwm/pwm-bcm2835.c
> > @@ -28,6 +28,7 @@ struct bcm2835_pwm {
> >
> > struct device *dev;
> > void __iomem *base;
> > struct clk *clk;
> >
> > + unsigned long rate;
> >
> > };
> >
> > static inline struct bcm2835_pwm *to_bcm2835_pwm(struct pwm_chip *chip)
> >
> > @@ -63,17 +64,11 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip,
> > struct pwm_device *pwm,>
> > {
> >
> > struct bcm2835_pwm *pc = to_bcm2835_pwm(chip);
> >
> > - unsigned long rate = clk_get_rate(pc->clk);
> >
> > unsigned long long period_cycles;
> > u64 max_period;
> >
> > u32 val;
> >
> > - if (!rate) {
> > - dev_err(pc->dev, "failed to get clock rate\n");
> > - return -EINVAL;
> > - }
> > -
> >
> > /*
> >
> > * period_cycles must be a 32 bit value, so period * rate /
> > NSEC_PER_SEC
> > * must be <= U32_MAX. As U32_MAX * NSEC_PER_SEC < U64_MAX the
> >
> > @@ -88,13 +83,13 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip,
> > struct pwm_device *pwm,>
> > * <=> period < ((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate
> > * <=> period <= ceil((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) /
rate)
> > - 1
> > */
> >
> > - max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC +
NSEC_PER_SEC
> > / 2, rate) - 1; + max_period = DIV_ROUND_UP_ULL((u64)U32_MAX *
> > NSEC_PER_SEC + NSEC_PER_SEC / 2, pc->rate) - 1;>
> > if (state->period > max_period)
> >
> > return -EINVAL;
> >
> > /* set period */
> >
> > - period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * rate,
> > NSEC_PER_SEC); + period_cycles = DIV_ROUND_CLOSEST_ULL(state->period *
> > pc->rate, NSEC_PER_SEC);>
> > /* don't accept a period that is too small */
> > if (period_cycles < PERIOD_MIN)
> >
> > @@ -103,7 +98,7 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip,
> > struct pwm_device *pwm,>
> > writel(period_cycles, pc->base + PERIOD(pwm->hwpwm));
> >
> > /* set duty cycle */
> >
> > - val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * rate, NSEC_PER_SEC);
> > + val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * pc->rate,
NSEC_PER_SEC);
> >
> > writel(val, pc->base + DUTY(pwm->hwpwm));
> >
> > /* set polarity */
> >
> > @@ -129,6 +124,7 @@ static const struct pwm_ops bcm2835_pwm_ops = {
> >
> > .request = bcm2835_pwm_request,
> > .free = bcm2835_pwm_free,
> > .apply = bcm2835_pwm_apply,
> >
> > + .atomic = true,
> >
> > .owner = THIS_MODULE,
> >
> > };
> >
> > @@ -156,6 +152,13 @@ static int bcm2835_pwm_probe(struct platform_device
> > *pdev)>
> > if (ret)
> >
> > return ret;
> >
> > + pc->rate = clk_get_rate(pc->clk);
> > + if (!pc->rate) {
> > + dev_err(pc->dev, "failed to get clock rate\n");
> > + ret = -EINVAL;
> > + goto add_fail;
> > + }
> > +
> >
> > pc->chip.dev = &pdev->dev;
> > pc->chip.ops = &bcm2835_pwm_ops;
> > pc->chip.npwm = 2;
--
TQ-Systems GmbH | M?hlstra?e 2, Gut Delling | 82229 Seefeld, Germany
Amtsgericht M?nchen, HRB 105018
Gesch?ftsf?hrer: Detlef Schneider, R?diger Stahl, Stefan Schneider
http://www.tq-group.com/
Hi Alexander,
Am 13.10.23 um 13:13 schrieb Alexander Stein:
> Hi,
>
> Am Freitag, 13. Oktober 2023, 13:04:48 CEST schrieb Stefan Wahren:
>> Hi Sean,
>>
>> Am 13.10.23 um 12:46 schrieb Sean Young:
>>> clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
>>> an rpi, simply fetch it once.
>> does it mean you checked all possible SoCs (BCM2835, BCM2836, BCM2837,
>> BCM2711, BCM2712) for this change?
>>
>> Is it impossible that the real clock can never be influenced by turbo
>> mode like SPI?
> Assuming the clock can change, which I would, then a clock notifier seems
> appropriate. See [1] for an example.
i remember a similiar approach for the CPU frequency for the RPi. At end
the we decided to let the firmware handle it and don't use clock
notifier, see [2] and the related links for more background. The fact
that the VideoCore has the real control makes it hard.
I don't want to say that's impossible.
[2] -
https://patchwork.kernel.org/project/linux-arm-kernel/cover/[email protected]/
>
> Best regards,
> Alexander
>
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/
> commit/?id=90ad2cbe88c22d0215225ab9594eeead0eb24fde
>
>> Best regards
>>
>>> Signed-off-by: Sean Young <[email protected]>
>>> ---
>>>
>>> drivers/pwm/pwm-bcm2835.c | 21 ++++++++++++---------
>>> 1 file changed, 12 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/drivers/pwm/pwm-bcm2835.c b/drivers/pwm/pwm-bcm2835.c
>>> index bdfc2a5ec0d6..59ea154dd657 100644
>>> --- a/drivers/pwm/pwm-bcm2835.c
>>> +++ b/drivers/pwm/pwm-bcm2835.c
>>> @@ -28,6 +28,7 @@ struct bcm2835_pwm {
>>>
>>> struct device *dev;
>>> void __iomem *base;
>>> struct clk *clk;
>>>
>>> + unsigned long rate;
>>>
>>> };
>>>
>>> static inline struct bcm2835_pwm *to_bcm2835_pwm(struct pwm_chip *chip)
>>>
>>> @@ -63,17 +64,11 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip,
>>> struct pwm_device *pwm,>
>>> {
>>>
>>> struct bcm2835_pwm *pc = to_bcm2835_pwm(chip);
>>>
>>> - unsigned long rate = clk_get_rate(pc->clk);
>>>
>>> unsigned long long period_cycles;
>>> u64 max_period;
>>>
>>> u32 val;
>>>
>>> - if (!rate) {
>>> - dev_err(pc->dev, "failed to get clock rate\n");
>>> - return -EINVAL;
>>> - }
>>> -
>>>
>>> /*
>>>
>>> * period_cycles must be a 32 bit value, so period * rate /
>>> NSEC_PER_SEC
>>> * must be <= U32_MAX. As U32_MAX * NSEC_PER_SEC < U64_MAX the
>>>
>>> @@ -88,13 +83,13 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip,
>>> struct pwm_device *pwm,>
>>> * <=> period < ((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate
>>> * <=> period <= ceil((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) /
> rate)
>>> - 1
>>> */
>>>
>>> - max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC +
> NSEC_PER_SEC
>>> / 2, rate) - 1; + max_period = DIV_ROUND_UP_ULL((u64)U32_MAX *
>>> NSEC_PER_SEC + NSEC_PER_SEC / 2, pc->rate) - 1;>
>>> if (state->period > max_period)
>>>
>>> return -EINVAL;
>>>
>>> /* set period */
>>>
>>> - period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * rate,
>>> NSEC_PER_SEC); + period_cycles = DIV_ROUND_CLOSEST_ULL(state->period *
>>> pc->rate, NSEC_PER_SEC);>
>>> /* don't accept a period that is too small */
>>> if (period_cycles < PERIOD_MIN)
>>>
>>> @@ -103,7 +98,7 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip,
>>> struct pwm_device *pwm,>
>>> writel(period_cycles, pc->base + PERIOD(pwm->hwpwm));
>>>
>>> /* set duty cycle */
>>>
>>> - val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * rate, NSEC_PER_SEC);
>>> + val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * pc->rate,
> NSEC_PER_SEC);
>>> writel(val, pc->base + DUTY(pwm->hwpwm));
>>>
>>> /* set polarity */
>>>
>>> @@ -129,6 +124,7 @@ static const struct pwm_ops bcm2835_pwm_ops = {
>>>
>>> .request = bcm2835_pwm_request,
>>> .free = bcm2835_pwm_free,
>>> .apply = bcm2835_pwm_apply,
>>>
>>> + .atomic = true,
>>>
>>> .owner = THIS_MODULE,
>>>
>>> };
>>>
>>> @@ -156,6 +152,13 @@ static int bcm2835_pwm_probe(struct platform_device
>>> *pdev)>
>>> if (ret)
>>>
>>> return ret;
>>>
>>> + pc->rate = clk_get_rate(pc->clk);
>>> + if (!pc->rate) {
>>> + dev_err(pc->dev, "failed to get clock rate\n");
>>> + ret = -EINVAL;
>>> + goto add_fail;
>>> + }
>>> +
>>>
>>> pc->chip.dev = &pdev->dev;
>>> pc->chip.ops = &bcm2835_pwm_ops;
>>> pc->chip.npwm = 2;
>
Hello,
On Fri, Oct 13, 2023 at 01:13:50PM +0200, Alexander Stein wrote:
> Am Freitag, 13. Oktober 2023, 13:04:48 CEST schrieb Stefan Wahren:
> > Am 13.10.23 um 12:46 schrieb Sean Young:
> > > clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
> > > an rpi, simply fetch it once.
> >
> > does it mean you checked all possible SoCs (BCM2835, BCM2836, BCM2837,
> > BCM2711, BCM2712) for this change?
> >
> > Is it impossible that the real clock can never be influenced by turbo
> > mode like SPI?
>
> Assuming the clock can change, which I would, then a clock notifier seems
> appropriate. See [1] for an example.
I'm not a fan. If the clock changes, the output also changes. With a
clock notifier you can soften the issue and reconfigure to something
similar as the original wave form, but a glitch happens for sure.
I already toyed with the thought to add clk_rate_exclusive_get() to all
PWM drivers, but didn't come around it yet.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-K?nig |
Industrial Linux Solutions | https://www.pengutronix.de/ |
On 13.10.23 г. 20:51 ч., Uwe Kleine-König wrote:
> Hello,
>
> On Fri, Oct 13, 2023 at 01:13:50PM +0200, Alexander Stein wrote:
>> Am Freitag, 13. Oktober 2023, 13:04:48 CEST schrieb Stefan Wahren:
>>> Am 13.10.23 um 12:46 schrieb Sean Young:
>>>> clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
>>>> an rpi, simply fetch it once.
>>>
>>> does it mean you checked all possible SoCs (BCM2835, BCM2836, BCM2837,
>>> BCM2711, BCM2712) for this change?
>>>
>>> Is it impossible that the real clock can never be influenced by turbo
>>> mode like SPI?
>>
>> Assuming the clock can change, which I would, then a clock notifier seems
>> appropriate. See [1] for an example.
>
> I'm not a fan. If the clock changes, the output also changes. With a
> clock notifier you can soften the issue and reconfigure to something
> similar as the original wave form, but a glitch happens for sure.
>
Right, but without notifier, everything rate related after the change
will be wrong
Ivo
Hello Ivaylo,
On Sat, Oct 14, 2023 at 09:51:12AM +0300, Ivaylo Dimitrov wrote:
> On 13.10.23 г. 20:51 ч., Uwe Kleine-König wrote:
> > Hello,
> >
> > On Fri, Oct 13, 2023 at 01:13:50PM +0200, Alexander Stein wrote:
> > > Am Freitag, 13. Oktober 2023, 13:04:48 CEST schrieb Stefan Wahren:
> > > > Am 13.10.23 um 12:46 schrieb Sean Young:
> > > > > clk_get_rate() may do a mutex lock. Since the clock rate cannot change on
> > > > > an rpi, simply fetch it once.
> > > >
> > > > does it mean you checked all possible SoCs (BCM2835, BCM2836, BCM2837,
> > > > BCM2711, BCM2712) for this change?
> > > >
> > > > Is it impossible that the real clock can never be influenced by turbo
> > > > mode like SPI?
> > >
> > > Assuming the clock can change, which I would, then a clock notifier seems
> > > appropriate. See [1] for an example.
> >
> > I'm not a fan. If the clock changes, the output also changes. With a
> > clock notifier you can soften the issue and reconfigure to something
> > similar as the original wave form, but a glitch happens for sure.
> >
>
> Right, but without notifier, everything rate related after the change will
> be wrong
So we agree clk_rate_exclusive_get() is the way to go?! It's simple, no
need for a notifier and no glitches.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | https://www.pengutronix.de/ |