On 04/05/11(Wed) 20:29, Miod Vallat wrote: > > > Speaking of DELAY()... it is implemented using the processor internal > > > counter register. Is this register impacted by frequency changes? If so, > > > shouldn't you update the computed ns_per_tick delay() constant? > > > > Reading the doc again, it's said that the time base register is clocked > > at one-fourth of the bus clock. But the DFS feature divides the processor > > to system bus ratio. So, if I understand well there is no impact on the > > time base counter frequency. > > Good. This is easy to check, does ntpd start complaining after running a > few minutes at `setperf=0' speed?
It doesn't complain and adjusts the clock the same way than with setperf=100. > > > Index: sys/arch/macppc/dev/dfs.c > > > +#include <sys/param.h> > > +#include <sys/filedesc.h> > > Could you use <sys/proc.h> instead of <sys/filedesc.h> here? This is the > preferred (yet objectionable) form of satisfying <sys/sysctl.h> > dependencies. Understood. > > +struct cfattach dfs_ca = { > > + sizeof(struct device), dfs_match, dfs_attach > > This needs to be sizeof(struct dfs_softc) now. That is, unless you want > to get funny panics after dfs0 attaches. Of course. I'm waiting some more test from the powermac owners to know if their machine support DFS or not, before sending a new diff.