[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <YzefysLuituL+LSA@yury-laptop>
Date: Fri, 30 Sep 2022 19:02:50 -0700
From: Yury Norov <yury.norov@...il.com>
To: Valentin Schneider <vschneid@...hat.com>
Cc: linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
"David S . Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>,
Jakub Kicinski <kuba@...nel.org>,
Paolo Abeni <pabeni@...hat.com>,
Rasmus Villemoes <linux@...musvillemoes.dk>
Subject: Re: [PATCH 1/7] cpumask: fix checking valid cpu range
On Fri, Sep 30, 2022 at 06:04:08PM +0100, Valentin Schneider wrote:
[...]
> > next_cpu is a valid CPU number for all, but not for cpumask_next().
> > The warning is valid. If we are at the very last cpu, what for we look
> > for next?
> >
>
> Consider:
>
> nr_cpu_ids=4
>
> A)
> cpumask: 0.1.1.0
> CPU 0 1 2 3
> n ^
> result: nr_cpu_ids
>
> B)
> cpumask: 0.0.1.1
> CPU 0 1 2 3
> n ^
> result: nr_cpu_ids + WARN
>
> Both scenarios are identical from a user perspective: a valid CPU number
> was passed in (either from smp_processor_id() or from a previous call to
> cpumask_next*()), but there are no more bits set in the cpumask. There's no
> more CPUs to search for in both scenarios, but only one produces as WARN.
It seems I have to repeat it for the 3rd time.
cpumask_next() takes shifted cpu index. That's why cpumask_check()
must shift the index in the other direction to keep all that
checking logic consistent.
This is a bad design, and all users of cpumask_next() must be aware of
this pitfall.
[...]
> > Maybe we should consider nr_cpu_ids as a special valid index for
> > cpumask_check(), a sign of the end of an array. This would help to
> > silence many warnings, like this one. For now I'm leaning towards that
> > it's more a hack than a meaningful change.
> >
>
> I agree, we definitely want to warn for e.g.
>
> cpumask_set_cpu(nr_cpu_ids, ...);
>
> Could we instead make cpumask_next*() immediately return nr_cpu_ids when
> passed n=nr_cpu_ids-1?
This is what FIND_NEXT_BIT() does. If you're suggesting to silence the
warning - what for do we need it at all?
> Also, what about cpumask_next_wrap()? That uses cpumask_next() under the
> hood and is bound to warn when wrapping after n=nr_cpu_ids-1, I think.
I'm working on a fix for it. Hopefully will merge it in next window.
Thanks,
Yury
Powered by blists - more mailing lists