Hi Heiko,

We didn't need the 200 Hz PWM in Linux, so I didn't test it right away. I did now, and the problem isn't present there, which I found weird given the fact that it is indeed similar code. So I dug into the kernel: The devil is in the details. :)

Kernel: c = (unsigned long long)period_cycles * duty_ns;

U-boot: c = (unsigned long long)(*period_c * duty_ns);

The braces cause the casting to be performed after the (truncated) multiplication, while without braces, the casting is done upfront.

Thank you for your reply and giving attention to this small bug.

Kind regards, Brecht

-----Original Message----- From: Heiko Schocher [mailto:hs@denx.de] Sent: dinsdag 28 april 2015 7:32 To: NEYRINCK Brecht Cc: u-boot@lists.denx.de Subject: Re: [U-Boot] [PATCH] bugfix i.mx6 pwm: prevent overflow of period_c * duty_ns by casting duty_ns to ull first. This bug came up when trying to create a 200 Hz PWM.

Hello Brecht Neyrinck,

Am 27.04.2015 14:11, schrieb Brecht Neyrinck:

---

drivers/pwm/pwm-imx-util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) mode change 100644 => 100755 drivers/pwm/pwm-imx-util.c

diff --git a/drivers/pwm/pwm-imx-util.c b/drivers/pwm/pwm-imx-util.c index f1d0b35..777a8bf 100644 --- a/drivers/pwm/pwm-imx-util.c +++ b/drivers/pwm/pwm-imx-util.c @@ -56,7 +56,7 @@ int pwm_imx_get_parms(int period_ns, int duty_ns, unsigned long *period_c, *prescale = *period_c / 0x10000 + 1;

*period_c /= *prescale;

• c = (unsigned long long)(*period_c * duty_ns);

• c = *period_c * (unsigned long long) duty_ns;

Thanks for this ... Hmm... this code is directly from linux drivers/pwm/pwm-imx.c ... Do you have running a linux on your hw? Could you verify this in linux too?

Thanks!

Acked-by: Heiko Schocher hs@denx.de

bye, Heiko

do_div(c, period_ns); *duty_c = c;