Re: [PATCH v3 2/2] vfs: avoid duplicating creds in faccessat if possible

From: Linus Torvalds <torvalds@linux-foundation.org>
To: Yury Norov <yury.norov@gmail.com>
Cc: Mateusz Guzik <mjguzik@gmail.com>,
	Alexander Potapenko <glider@google.com>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Kees Cook <keescook@chromium.org>,
	Eric Biggers <ebiggers@google.com>,
	Christian Brauner <brauner@kernel.org>,
	serge@hallyn.com, paul@paul-moore.com,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-security-module@vger.kernel.org
Subject: Re: [PATCH v3 2/2] vfs: avoid duplicating creds in faccessat if possible
Date: Sat, 4 Mar 2023 11:02:36 -0800	[thread overview]
Message-ID: <CAHk-=whJSXQfPGjpj8SAgEn5bOHOGvHPvFQM6GutcUU7kS8qXw@mail.gmail.com> (raw)
In-Reply-To: <ZALcbQoKA7K8k2gJ@yury-laptop>

On Fri, Mar 3, 2023 at 9:51 PM Yury Norov <yury.norov@gmail.com> wrote:
>
> At that time you was OK with CONFIG_FORCE_NR_CPUS, only suggested to
> hide it behind CONFIG_EXPERT:

I think there was a mis-communication.

I as violently *not* ok with that question at all. I think our Kconfig
phase is really annoying and nasty, and we should not ask people
questions that they don't know what they mean.

So putting it behind EXPERT was a "at that point, the question is
gone", and I'm ok with the config variable existing.

But..

I am *not* ok with you then thinking that "oh, the config variable
exists, so our default code generation can be complete crap".

The kernel should do well by default. No amount of "but you could go
into magic config variables and then force options that might be ok
for embedded systems to make it generate ok code".

I think it's completely crazy that the distros enable MAXSMP. But
that's their choice. A *sane* distro should not do that, and then we
limit the normal kernel to something sane like a couple of hundreds of
CPUs rather than thousands of them (and the associated huge overhead).

At that point, things like cpumap_clear() should be a couple of stores
- not a call-out to a variable-sized memset().

Notice? Simple and understandable Kconfig questions like "do you
*really* need to support thousands of CPU's"

Not that FORCE_NR_CPUS that is _completely_ useless to any
distribution and thus completely unacceptable as a regular question.

See the difference?

          Linus