Re: Consider switching to WQ_UNBOUND messages (was: Re: [PATCH v2 6/7] workqueue: Report work funcs that trigger automatic CPU_INTENSIVE mechanism)

[Tejun Heo <tj@kernel.org>]

On Wed, Jul 12, 2023 at 02:27:45PM +0200, Peter Zijlstra wrote:
> On Wed, Jul 12, 2023 at 11:04:16AM +0200, Geert Uytterhoeven wrote:
> > Hoi Peter,
> > 
> > On Wed, Jul 12, 2023 at 10:05 AM Peter Zijlstra <peterz@infradead.org> wrote:
> > > On Tue, Jul 11, 2023 at 11:39:17AM -1000, Tejun Heo wrote:
> > > > I wonder whether the right thing to do here is somehow scaling the threshold
> > > > according to the relative processing power. It's difficult to come up with a
> > > > threshold which works well across the latest & fastest and really tiny CPUs.
> > > > I'll think about it some more but if you have some ideas, please feel free
> > > > to suggest.
> > >
> > > We could scale by BogoMIPS I suppose, it's a bogus measurement, as per
> > > the name, but it does have some relation to how fast the machine is.
> > 
> > That's gonna fail miserably on e.g. ARM and RISC-V, where BogoMIPS
> > depends on some timer frequency.
> > 
> > R-Car M2-W with 1.5 GHz Cortex-A15: 40.00 BogoMIPS
> > R-Car V4H with 1.8 GHz Cortex-A76: 33.33 BogoMIPS
> > 
> > while the real slow 48 MHz VexRiscV gets 128 BogoMIPS.
> 
> Hehe, OK, really bogus then. Lets file this idea in the bit-bucket then.

I think it can still be useful. On ryzen 3975wx, it's 6989.92, so while it
may be off by some hundreds of percents, there are still orders of magnitude
signal range and that should be enough to suppress most spurious warnings.
I'll post something later today.

Thanks.

-- 
tejun