Re: [PATCH -mm] Blackfin: blackfin i2c driver

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 7 Mar 2007 13:17:57 +0800 "Sonic Zhang" <sonic.adi@gmail.com> wrote:

> On 3/6/07, Andrew Morton <akpm@linux-foundation.org> wrote:
> > On Tue, 06 Mar 2007 14:54:18 +0800 "Wu, Bryan" <bryan.wu@analog.com> wrote:
> >
> > > Hi folks,
> > >
> > > [PATCH] Blackfin: blackfin i2c driver
> > >
> 
> > > +     struct i2c_msg *pmsg;
> > > +     int i, ret;
> > > +     int rc = 0;
> > > +
> > > +     if (!(bfin_read_TWI_CONTROL() & TWI_ENA))
> > > +             return -ENXIO;
> > > +
> > > +     down(&iface->twi_lock);
> > > +
> > > +     while (bfin_read_TWI_MASTER_STAT() & BUSBUSY) {
> > > +             up(&iface->twi_lock);
> > > +             schedule();
> > > +             down(&iface->twi_lock);
> > > +     }
> >
> > That's a busy loop until this task's timeslice has expired.  It'll work,
> > but it'll suck a bit.  (Repeated in several places)
> >
> 
> OK, I change it into yield(). So, current process will be move to the
> tail of the run queue. Is that OK with you?

Nope, yield is terribly bad when there are busy processes running: it can
stall for a very long time indeed,

Is this hardware not capable of generating an interrupt when BUSBUSY gets
negated?

I guess not, in which case you're stuck with having to poll it - probably
use a cond_resched() in the loop, and an angry comment.