Message-ID: <20170105105329.5e75291d@bbrezillon>
Date:   Thu, 5 Jan 2017 10:53:29 +0100
From:   Boris Brezillon <boris.brezillon@...e-electrons.com>
To:     Lukasz Majewski <lukma@...x.de>
Cc:     Thierry Reding <thierry.reding@...il.com>,
        Sascha Hauer <s.hauer@...gutronix.de>,
        Stefan Agner <stefan@...er.ch>, linux-pwm@...r.kernel.org,
        Bhuvanchandra DV <bhuvanchandra.dv@...adex.com>,
        linux-kernel@...r.kernel.org,
        Lothar Wassmann <LW@...o-electronics.de>,
        kernel@...gutronix.de, Fabio Estevam <fabio.estevam@....com>,
        Lukasz Majewski <l.majewski@...ess.pl>
Subject: Re: [PATCH v4 07/11] pwm: imx: Provide atomic PWM support for i.MX
 PWMv2

On Thu, 5 Jan 2017 10:35:05 +0100
Lukasz Majewski <lukma@...x.de> wrote:

> On Thu, 5 Jan 2017 10:19:35 +0100
> Boris Brezillon <boris.brezillon@...e-electrons.com> wrote:
> 
> > On Thu, 5 Jan 2017 10:03:47 +0100
> > Lukasz Majewski <lukma@...x.de> wrote:
> >   
> > > On Thu, 5 Jan 2017 09:50:35 +0100
> > > Boris Brezillon <boris.brezillon@...e-electrons.com> wrote:
> > >   
> > > > On Thu,  5 Jan 2017 00:36:50 +0100
> > > > Lukasz Majewski <lukma@...x.de> wrote:
> > > >     
> > > > > This commit provides apply() callback implementation for i.MX's
> > > > > PWMv2.
> > > > > 
> > > > > Suggested-by: Stefan Agner <stefan@...er.ch>
> > > > > Suggested-by: Boris Brezillon
> > > > > <boris.brezillon@...e-electrons.com> Signed-off-by: Lukasz
> > > > > Majewski <l.majewski@...ess.pl> Reviewed-by: Boris Brezillon
> > > > > <boris.brezillon@...e-electrons.com> ---
> > > > > Changes for v4:
> > > > > - Avoid recalculation of PWM parameters when disabling PWM
> > > > > signal
> > > > > - Unconditionally call clk_prepare_enable(imx->clk_per) and
> > > > >   clk_disable_unprepare(imx->clk_per)    
> > > > 
> > > > I don't see that one, but I'm not sure this is actually needed.
> > > > In the enabled path we enable the clk before accessing the
> > > > registers, and in the disable path, assuming you change the code
> > > > according to my suggestion, the clk should already be enabled
> > > > when you write to MX3_PWMCR.
> > > >     
> > > > > 
> > > > > Changes for v3:
> > > > > - Remove ipg clock enable/disable functions
> > > > > 
> > > > > Changes for v2:
> > > > > - None
> > > > > ---
> > > > >  drivers/pwm/pwm-imx.c | 67
> > > > > +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file
> > > > > changed, 67 insertions(+)
> > > > > 
> > > > > diff --git a/drivers/pwm/pwm-imx.c b/drivers/pwm/pwm-imx.c
> > > > > index 60cdc5c..134dd66 100644
> > > > > --- a/drivers/pwm/pwm-imx.c
> > > > > +++ b/drivers/pwm/pwm-imx.c
> > > > > @@ -249,6 +249,72 @@ static int imx_pwm_config(struct pwm_chip
> > > > > *chip, return ret;
> > > > >  }
> > > > >  
> > > > > +static int imx_pwm_apply_v2(struct pwm_chip *chip, struct
> > > > > pwm_device *pwm,
> > > > > +			    struct pwm_state *state)
> > > > > +{
> > > > > +	unsigned long period_cycles, duty_cycles, prescale;
> > > > > +	struct imx_chip *imx = to_imx_chip(chip);
> > > > > +	struct pwm_state cstate;
> > > > > +	unsigned long long c;
> > > > > +	int ret;
> > > > > +
> > > > > +	pwm_get_state(pwm, &cstate);
> > > > > +
> > > > > +	if (state->enabled) {
> > > > > +		c = clk_get_rate(imx->clk_per);
> > > > > +		c *= state->period;
> > > > > +
> > > > > +		do_div(c, 1000000000);
> > > > > +		period_cycles = c;
> > > > > +
> > > > > +		prescale = period_cycles / 0x10000 + 1;
> > > > > +
> > > > > +		period_cycles /= prescale;
> > > > > +		c = (unsigned long long)period_cycles *
> > > > > state->duty_cycle;
> > > > > +		do_div(c, state->period);
> > > > > +		duty_cycles = c;
> > > > > +
> > > > > +		/*
> > > > > +		 * according to imx pwm RM, the real period
> > > > > value should be
> > > > > +		 * PERIOD value in PWMPR plus 2.
> > > > > +		 */
> > > > > +		if (period_cycles > 2)
> > > > > +			period_cycles -= 2;
> > > > > +		else
> > > > > +			period_cycles = 0;
> > > > > +    
> > > > 
> > > > Starting here...
> > > >     
> > > > > +		ret = clk_prepare_enable(imx->clk_per);
> > > > > +		if (ret)
> > > > > +			return ret;
> > > > > +
> > > > > +		/*
> > > > > +		 * Wait for a free FIFO slot if the PWM is
> > > > > already enabled, and
> > > > > +		 * flush the FIFO if the PWM was disabled and
> > > > > is about to be
> > > > > +		 * enabled.
> > > > > +		 */
> > > > > +		if (cstate.enabled)
> > > > > +			imx_pwm_wait_fifo_slot(chip, pwm);
> > > > > +		else if (state->enabled)
> > > > > +			imx_pwm_sw_reset(chip);    
> > > > 
> > > > ... till here, should be replaced by:
> > > > 
> > > > 
> > > > 		/*
> > > > 		 * Wait for a free FIFO slot if the PWM is already
> > > > enabled, and
> > > > 		 * flush the FIFO if the PWM was disabled and is
> > > > about to be
> > > > 		 * enabled.
> > > > 		 */
> > > > 		if (cstate.enabled) {
> > > > 			imx_pwm_wait_fifo_slot(chip, pwm);
> > > > 		} else {
> > > > 			ret = clk_prepare_enable(imx->clk_per);
> > > > 			if (ret)
> > > > 				return ret;    
> > > 
> > > In the other mail you mentioned that the clock should be enabled
> > > unconditionally to fix issue on iMX7 when we want to access
> > > registers.  
> > 
> > When I said unconditionally, I meant call
> > clk_prepare_enable(imx->clk_per) just after pwm_get_state() (at the
> > beginning of the function) and call clk_disable_(imx->clk_per) just
> > before returning 0 at the end of the function. This way you're
> > guaranteed that the clk is always enabled when you access registers in
> > between. Of course, you still need the the clk_prepare_enable() and
> > clk_disable_unprepare() in the enable and disable path to keep the clk
> > enabled when the PWM is enabled, and to disable it when the clk is
> > being disabled.  
> 
> Why you didn't said this when I pasted pseudo code to the other e-mail
> and explicitly asked if this is correct? By doing that I wanted to
> avoid situations like this (so I will have to repost this again and
> again .....).

Because I thought the unconditional clk_enable/disable was not part of
your example and would be added later, and I must admit I didn't
carefully read the code chunk. Anyway, it's not a big deal, you're only
at v4.

> 
> 
> > 
> > But, while reviewing your patch I realized this was actually unneeded
> > (see the explanation in my previous review).
> >   
> > > 
> > > Now it depends on cstate.enabled flag. 
> > > 
> > > So we end up with 
> > > 
> > > if (state.enabled && !cstate.enabled)
> > > 	clk_preapre_enable();  
> > 
> > Yep, and that's correct.  
> 
> And following patch:
> http://patchwork.ozlabs.org/patch/709510/
> 
> address this issue.

Yes, that was needed because the enable/disable path were not
separated, and we were unconditionally writing to the IP registers
even when the PWM was already disabled (which is probably the case
generating the fault reported by Stefan). This is not the case anymore,
but let's wait for Stefan to confirm this.

Also note that Stefan's patch is introducing an 'unbalanced clk
prepapre/enable refcount' bug if ->apply() is called several times
consecutively with the same state->enabled value.

> 
> >   
> > > 
> > > which was the reason of iMX7 failure reported by Stefan to v3.  
> > 
> > Stefan, do you confirm that? I don't see how this can possibly fail
> > since the clk is either already enabled (cstate.enabled) or we enable
> > it (!cstate.enable) before accessing registers.  
> 
> http://patchwork.ozlabs.org/patch/709510/
> 
> > 
> > What's for sure is that your implementation is introducing possible
> > unbalanced ref counting on the per clk.  
> 
> You mean when somebody call twice ->apply() with "state->enabled".

Yep, when someone calls twice ->apply() with state->enabled set to true
or twice with state->enabled set to false.

Powered by blists - more mailing lists