Re: [PATCH] lib/test_free_pages: Add basic progress indicators

[Geert Uytterhoeven]

Hi Matthew,

On Mon, Oct 19, 2020 at 4:05 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> On Sun, Oct 18, 2020 at 08:12:52PM +0300, Mike Rapoport wrote:
> > On Sun, Oct 18, 2020 at 04:01:46PM +0100, Matthew Wilcox wrote:
> > > On Sun, Oct 18, 2020 at 04:39:27PM +0200, Geert Uytterhoeven wrote:
> > > > On Sun, Oct 18, 2020 at 4:25 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> > > > > On Sun, Oct 18, 2020 at 04:04:45PM +0200, Geert Uytterhoeven wrote:
> > > > > > The test module to check that free_pages() does not leak memory does not
> > > > > > provide any feedback whatsoever its state or progress, but may take some
> > > > > > time on slow machines.  Add the printing of messages upon starting each
> > > > > > phase of the test, and upon completion.
> > > > >
> > > > > It's not supposed to take a long time.  Can you crank down that 1000 *
> > > >
> > > > It took 1m11s on ARAnyM, running on an i7-8700K.
> > > > Real hardware may even take longer.
> > >
> > > 71 seconds is clearly too long.  0.7 seconds would be fine, so 10 * 1000
> > > would be appropriate, but then that's only 320MB which might not be
> > > enough to notice on a modern machine.
> > >
> > > > > 1000 to something more appropriate?
> > > >
> > > > What would be a suitable value? You do want to see it "leak gigabytes
> > > > of memory and probably OOM your system" if something's wrong,
> > > > so decreasing the value a lot may not be a good idea?
> > > >
> > > > Regardless, if it OOMs, I think you do want to see this happens
> > > > while running this test.
> > >
> > > How about scaling with the amount of memory on the machine?
> > >
> > > This might cause problems on machines with terabytes of memory.
> > > Maybe we should cap it at a terabyte?
> >
> > On ARAnyM wih 782 MBytes of RAM running on i7-8650U it takes ~1.75
> > seconds.
>
> That seems like a somewhat unusual configuration.  I think it's pretty
> strange to find an actual m68k with more than 128MB of memory.  I mean,
> I can set up my laptop to believe it has 64TB of memory, and this will
> run slowly, but I don't think it's any real problem.

Have you tried it on one of your parisc boxes? They tend to have quite
some RAM, compared to CPU speed.

I gave the unmodified test a try on a 1.5 GHz Cortex-A15, which is still
a decent arm32 system, and it took 4.25s.

> > Still, I think adding some verbosity to the test wouldn't hurt ;-)
>
> I prefer the unix philosophy of only emitting messages if something's
> wrong.

This is not a normal program, but a test.
Alternatively, convert it to kunit, so we'll at least be aware when
it starts running, and when it has completed ;-)

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds