Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 2 Aug 2023 17:48:29 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Mike Rapoport <rppt@kernel.org>
Cc:     yury.norov@gmail.com, andriy.shevchenko@linux.intel.com,
        linux@rasmusvillemoes.dk,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        tglx@linutronix.de
Subject: Re: [PATCH 2/2] mm,nodemask: Use nr_node_ids
Message-ID: <20230802154829.GB214207@hirez.programming.kicks-ass.net>
References: <20230802112458.230221601@infradead.org>
 <20230802112525.633758009@infradead.org>
 <20230802153251.GF2607694@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230802153251.GF2607694@kernel.org>
Precedence: bulk

On Wed, Aug 02, 2023 at 06:32:51PM +0300, Mike Rapoport wrote:
> > +/*
> > + * We have several different "preferred sizes" for the nodemask
> > + * operations, depending on operation.
> > + *
> > + * For example, the bitmap scanning and operating operations have
> > + * optimized routines that work for the single-word case, but only when
> > + * the size is constant. So if NR_CPUS fits in one single word, we are
> 
>                                 ^ MAX_NUMNODES?
> 
> > + * better off using that small constant, in order to trigger the
> > + * optimized bit finding. That is 'small_nodemask_size'.
> > + *
> > + * The clearing and copying operations will similarly perform better
> > + * with a constant size, but we limit that size arbitrarily to four
> > + * words. We call this 'large_nodemask_size'.
> > + *
> > + * Finally, some operations just want the exact limit, either because
> > + * they set bits or just don't have any faster fixed-sized versions. We
> > + * call this just 'nr_nodemask_bits'.
> > + *
> > + * Note that these optional constants are always guaranteed to be at
> > + * least as big as 'nr_node_ids' itself is, and all our nodemask
> > + * allocations are at least that size (see nodemask_size()). The
> 
> We don't have nodemask_size(). NODEMASK_ALLOC() actually allocates memory
> only when NODE_SHIFT > 8 and it always uses the static size.
> 
> > + * optimization comes from being able to potentially use a compile-time
> > + * constant instead of a run-time generated exact number of CPUs.
> 
>                                                               ^ nodes?

Durr, clearly I didn't actually read the comment after I 'borrowed' it
and regex'ed it into 'shape'.

I'll go fix, thanks!