Hi Alexey,
Alexey Romanov (AVRomanov@xxxxxxxxxxxxxx) wrote:
> > > +static int meson_rng_wait_status(void __iomem *cfg_addr, int bit)
> > > +{
> > > + u32 status;
> > > + u32 cnt = 0;
> > > +
> > > + do {
> > > + status = readl_relaxed(cfg_addr) & BIT(bit);
> > > + cpu_relax();
> > > + } while (status && (cnt++ < RETRY_CNT));
> > > +
> > Could you use readl_relaxed_poll_timeout here instead of open coding the
> > loop?
> At first I also thought about this API. But later I came to the
> conclusion that it is inappropriate here:
> 1. We can't call rng_read from an atomic context.
Agreed. A hwrng read function may sleep (an example for this is
exynos_trng_do_read). But this doesn't prevent us from using
readl_relaxed_poll_timeout.
> 2. RNG for me looks like a very lightweight primitive to me that
> should work quiclky.
> But, now I looked again at the API and realized that we can use
> readl_relaxed_poll_timeout_atomic() instead of
> readl_relaxed_poll_timeout(). What do you think?
Ok, if you know that your rng hardware won't need much time to set the
bit that you're checking, you may use readl_relaxed_poll_timeout_atomic.
stm32_rtc_set_alarm does something similar.
Best regards,
Martin
