At Thu, 3 Sep 2009 17:26:18 +0200, Giovanni Maruzzelli wrote:
On Thu, Sep 3, 2009 at 5:12 PM, Giovanni Maruzzelligmaruzz.lists@gmail.com wrote:
I'll do it and report
I put hrtimer=0
I put the printk immediately after the spin_unlock, just before return, this is a some second run (gave me bad behavior, as before, maybe worst):
Sep 3 17:20:20 localhost kernel: giovanni line: 291, dpcm->frac_pos / HZ=0, jiffies=91837395 Sep 3 17:20:20 localhost kernel: giovanni line: 291, dpcm->frac_pos / HZ=48, jiffies=91837398 Sep 3 17:20:20 localhost kernel: giovanni line: 291, dpcm->frac_pos / HZ=48, jiffies=91837398 Sep 3 17:20:20 localhost kernel: giovanni line: 291, dpcm->frac_pos / HZ=80, jiffies=91837400 Sep 3 17:20:20 localhost kernel: giovanni line: 291, dpcm->frac_pos / HZ=80, jiffies=91837400
So, this is a kernel with the variable HZ. That's the whole problem. My patch also doesn't take it into account but assumes the constant HZ. We'll need to convert the stuff to msecs_to_jiffies() and co...
Takashi