Re: [PATCH v3] clocksource: Scale the max retry number of watchdog read according to CPU numbers

[Feng Tang]

On Tue, Feb 20, 2024 at 07:24:27AM -0800, Paul E. McKenney wrote:
> On Mon, Feb 19, 2024 at 09:20:31PM -0500, Waiman Long wrote:
> > 
> > On 2/19/24 09:37, Feng Tang wrote:
> > > Hi Thomas,
> > > 
> > > On Mon, Feb 19, 2024 at 12:32:05PM +0100, Thomas Gleixner wrote:
> > > > On Mon, Jan 29 2024 at 21:45, Feng Tang wrote:
> > > > > +static inline long clocksource_max_watchdog_read_retries(void)
> > > > > +{
> > > > > +	long max_retries = max_cswd_read_retries;
> > > > > +
> > > > > +	if (max_cswd_read_retries <= 0) {
> > > > > +		/* santity check for user input value */
> > > > > +		if (max_cswd_read_retries != -1)
> > > > > +			pr_warn_once("max_cswd_read_retries was set with an invalid number: %ld\n",
> > > > > +				max_cswd_read_retries);
> > > > > +
> > > > > +		max_retries = ilog2(num_online_cpus()) + 1;
> > > > I'm getting tired of these knobs and the horrors behind them. Why not
> > > > simply doing the obvious:
> > > > 
> > > >         retries = ilog2(num_online_cpus()) + 1;
> > > > 
> > > > and remove the knob alltogether?
> > > Thanks for the suggestion! Yes, this makes sense to me. IIUC, the
> > > 'max_cswd_read_retries' was introduced mainly to cover different
> > > platforms' requirement, which could now be covered by the new
> > > self-adaptive number.
> > > 
> > > If there is no concern from other developers, I will send a new
> > > version in this direction.
> > 
> > I see no problem simplifying it.
> 
> My guess is that we will eventually end up with something like this:
> 
> 	retries = ilog2(num_online_cpus()) / 2 + 1;

Good point! Initially when writing the patch, I did try to search if
there is a 'ilog4' api :) as the ilog2 of that 8-socket machine is
about 10, which is more than enough.

Thanks,
Feng

> 
> but I am not at all opposed to starting without the division by 2.
> 
> 							Thanx, Paul