Hello,

On Tue, Jun 02, 2020 at 03:31:10PM -0700, Guru Das Srinagesh wrote:
> Since the PWM framework is switching struct pwm_state.period's
> datatype to u64, prepare for this transition by using
> DIV_ROUND_UP_ULL to handle a 64-bit dividend, and div64_u64 to handle a
> 64-bit divisor.
> 
> Also handle a possible overflow in the calculation of period_cycles when
> both clk_rate and period are large numbers. This logic was unit-tested
> out by calculating period_cycles using both the existing logic and the
> proposed one, and the results are as below.
> 
> ----------------------------------------------------------------------------
>  clk_rate            period           existing            proposed
> ----------------------------------------------------------------------------
> 1000000000   18446744073709551615    18446744072    18446744073000000000
>                    (U64_MAX)
> ----------------------------------------------------------------------------
> 1000000000        4294967291         4294967291         4294967291
> ----------------------------------------------------------------------------
> 
> Overflow occurs in the first case with the existing logic, whereas the
> proposed logic handles it better, sacrificing some precision in a best-effort
> attempt to handle overflow. As for the second case where there are
> more typical values of period, the proposed logic handles that correctly
> too.
> 
> Cc: Shawn Guo <shawnguo@kernel.org>
> Cc: Sascha Hauer <s.hauer@pengutronix.de>
> Cc: Pengutronix Kernel Team <kernel@pengutronix.de>
> Cc: Fabio Estevam <festevam@gmail.com>
> Cc: NXP Linux Team <linux-imx@nxp.com>
> Signed-off-by: Guru Das Srinagesh <gurus@codeaurora.org>
> ---
>  drivers/pwm/pwm-imx27.c | 48 +++++++++++++++++++++++++++++++++++++++++++-----
>  1 file changed, 43 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/pwm/pwm-imx27.c b/drivers/pwm/pwm-imx27.c
> index 732a6f3..147729d 100644
> --- a/drivers/pwm/pwm-imx27.c
> +++ b/drivers/pwm/pwm-imx27.c
> @@ -202,7 +202,7 @@ static void pwm_imx27_wait_fifo_slot(struct pwm_chip *chip,
>  	sr = readl(imx->mmio_base + MX3_PWMSR);
>  	fifoav = FIELD_GET(MX3_PWMSR_FIFOAV, sr);
>  	if (fifoav == MX3_PWMSR_FIFOAV_4WORDS) {
> -		period_ms = DIV_ROUND_UP(pwm_get_period(pwm),
> +		period_ms = DIV_ROUND_UP_ULL(pwm_get_period(pwm),
>  					 NSEC_PER_MSEC);
>  		msleep(period_ms);
>  
> @@ -212,6 +212,45 @@ static void pwm_imx27_wait_fifo_slot(struct pwm_chip *chip,
>  	}
>  }
>  
> +static int pwm_imx27_calc_period_cycles(const struct pwm_state *state,
> +					unsigned long clk_rate,
> +					unsigned long *period_cycles)
> +{
> +	u64 c = 0, c1, c2;
> +
> +	c1 = clk_rate;
> +	c2 = state->period;
> +	if (c2 > c1) {
> +		c2 = c1;
> +		c1 = state->period;
> +	}
> +
> +	if (!c1 || !c2) {
> +		pr_err("clk rate and period should be nonzero\n");
> +		return -EINVAL;
> +	}
> +
> +	if (c2 <= div64_u64(U64_MAX, c1)) {
> +		c = c1 * c2;
> +		do_div(c, 1000000000);
> +	} else if (c2 <= div64_u64(U64_MAX, div64_u64(c1, 1000))) {
> +		do_div(c1, 1000);
> +		c = c1 * c2;
> +		do_div(c, 1000000);
> +	} else if (c2 <= div64_u64(U64_MAX, div64_u64(c1, 1000000))) {
> +		do_div(c1, 1000000);
> +		c = c1 * c2;
> +		do_div(c, 1000);
> +	} else if (c2 <= div64_u64(U64_MAX, div64_u64(c1, 1000000000))) {
> +		do_div(c1, 1000000000);
> +		c = c1 * c2;

	} else {
		???

> +	}
> +
> +	*period_cycles = c;
> +
> +	return 0;
> +}
> +
>  static int pwm_imx27_apply(struct pwm_chip *chip, struct pwm_device *pwm,
>  			   const struct pwm_state *state)
>  {
> @@ -226,10 +265,9 @@ static int pwm_imx27_apply(struct pwm_chip *chip, struct pwm_device *pwm,
>  	pwm_get_state(pwm, &cstate);
>  
>  	clkrate = clk_get_rate(imx->clk_per);
> -	c = clkrate * state->period;
> -
> -	do_div(c, NSEC_PER_SEC);
> -	period_cycles = c;
> +	ret = pwm_imx27_calc_period_cycles(state, clkrate, &period_cycles);
> +	if (ret)
> +		return ret;

I would expect this problem to be generic enough that it justifies a
function in pwm-core (or even more generic?).

Something like:

	/*
	 * calculates *result = a * b / c.
	 * Returns false on overflow (and doesn't assign result in this
	 * case); true otherwise.
	 */
	static inline bool multdiv(u64 *result, u64 a, u64 b, u64 c)

I'd split this out in a separate patch though? (Or is imx27 the only
driver in this series with this problem?)

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |