On Monday 24 September 2012, Chunhe Lan wrote:
>     OK. As you have mentioned, it would been modified to such:
> 
> static inline void mmc_delay(unsigned int ms)
> {
>          if (ms < 1000 / HZ) {
>                  cond_resched();
>                  msleep(ms);
>          } else {
>                  msleep(ms);
>          }
> }

This version would be rather broken, because it compares times
in two different units (ms and jiffies), and because it
does a cond_resched() directly before an msleep: both of which
end up calling schedule() and being away for some time,
cond_resched() for an unknown time, and msleep for a minimum
time on top of that.

> OR such:
> 
> static inline void mmc_delay(unsigned int ms)
> {
>           msleep(ms);
> }

That would be my preferred choice, unless someone has specific issues with this.

> OR other code?

Well, in principle, you could implement something like

static inline void mmc_delay(unsigned int ms)
{
        ktime_t end = ktime_add_us(ktime_get(), ms * 1000);

        while (1) {
                s64 remaining;

                cond_resched();

                remaining = ktime_to_us(ktime_sub(end, ktime_get()));

                if (remaining < 0)
                        break;

                udelay(min_t(u32, remaining, 100));
        }
}

        Arnd
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Reply via email to