[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZbQQXO5YhKhdr1Ou@bombadil.infradead.org>
Date: Fri, 26 Jan 2024 12:04:44 -0800
From: Luis Chamberlain <mcgrof@...nel.org>
To: Matthew Wilcox <willy@...radead.org>
Cc: kernel test robot <oliver.sang@...el.com>,
Daniel Gomez <da.gomez@...sung.com>, oe-lkp@...ts.linux.dev,
lkp@...el.com, linux-kernel@...r.kernel.org,
"gost.dev@...sung.com" <gost.dev@...sung.com>,
Pankaj Raghav <p.raghav@...sung.com>
Subject: Re: [PATCH 1/2] test_xarray: add tests for advanced multi-index use
On Fri, Jan 26, 2024 at 07:26:17PM +0000, Matthew Wilcox wrote:
> On Fri, Jan 26, 2024 at 11:12:03AM -0800, Luis Chamberlain wrote:
> > On Fri, Nov 17, 2023 at 01:01:18PM -0800, Luis Chamberlain wrote:
> > > On Fri, Nov 17, 2023 at 08:58:05PM +0000, Matthew Wilcox wrote:
> > > > On Fri, Nov 17, 2023 at 12:54:09PM -0800, Luis Chamberlain wrote:
> > > > > +/*
> > > > > + * Can be used in contexts which busy loop on large number of entries but can
> > > > > + * sleep and timing is if no importance to test correctness.
> > > > > + */
> > > > > +#define XA_BUG_ON_RELAX(xa, x) do { \
> > > > > + if ((tests_run % 1000) == 0) \
> > > > > + schedule(); \
> > > > > + XA_BUG_ON(xa, x); \
> > > > > +} while (0)
> > > >
> > > > That is awful. Please don't do that. You're mixing two completely
> > > > unrelated thing into the same macro, which makes no sense. Not only
> > > > that, it's a macro which refers to something in the containing
> > > > environment that isn't a paramter to the macro.
> > >
> > > I figured you'd puke. Would you prefer I just open code the check on the loop
> > > though? I'm sure another alternative is we *not care* about these
> > > overloaded systems running the test. What would you prefer?
> >
> > OK without any particular preferences outlined this is what I have,
> > splitting the two contexts and making the busy loop fix clearer.
> >
> > +#define XA_BUSY_LOOP_RELAX(xa, x) do { \
> > + if ((i % 1000) == 0) \
> > + schedule(); \
> > +} while (0)
> > +
> > +/*
> > + * Can be used in contexts which busy loop on large number of entries but can
> > + * sleep and timing is if no importance to test correctness.
> > + */
> > +#define XA_BUG_ON_RELAX(i, xa, x) do { \
> > + XA_BUSY_LOOP_RELAX(i); \
> > + XA_BUG_ON(xa, x); \
> > +} while (0)
>
> No. XA_BUG_ON_RELAX is not OK. Really.
>
> We have a perfectly good system for "relaxing":
>
> xas_for_each_marked(&xas, page, end, PAGECACHE_TAG_DIRTY) {
> xas_set_mark(&xas, PAGECACHE_TAG_TOWRITE);
> if (++tagged % XA_CHECK_SCHED)
> continue;
>
> xas_pause(&xas);
> xas_unlock_irq(&xas);
> cond_resched();
> xas_lock_irq(&xas);
> }
And yet we can get a soft lockup with order 20 (1,048,576 entries),
granted busy looping over 1 million entries is insane, but it seems it
the existing code may not be enough to avoid the soft lockup. Also
cond_resched() may be eventually removed [0].
Anyway, we can ignore the soft lockup then.
[0] https://lwn.net/Articles/950581/
Luis
Powered by blists - more mailing lists