Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200930065617.934638-1-palmerdabbelt@google.com>
 <20201002064905.GA27115@infradead.org> <5692716.lOV4Wx5bFT@positron.chronox.de>
 <CAMj1kXEnOh4MBiVVgkhd4P81eRPCVi3+y6JcD58jL45-eh324A@mail.gmail.com> <20201005061918.GB1856@infradead.org>
In-Reply-To: <20201005061918.GB1856@infradead.org>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Mon, 5 Oct 2020 08:24:46 +0200
Message-ID: <CAMj1kXFC=6G1pqFAcjASrKJfzguO3k9Rv-7NsmTaX8qNre5TGw@mail.gmail.com>
Subject: Re: [PATCH] crypto: jitterentropy - bind statically into kernel
To:     Christoph Hellwig <hch@infradead.org>
Cc:     "Stephan M??ller" <smueller@chronox.de>,
        Palmer Dabbelt <palmerdabbelt@google.com>,
        linux-riscv <linux-riscv@lists.infradead.org>,
        Android Kernel Team <kernel-team@android.com>,
        kernel test robot <lkp@intel.com>,
        Linux Crypto Mailing List <linux-crypto@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, 5 Oct 2020 at 08:19, Christoph Hellwig <hch@infradead.org> wrote:
>
> On Sun, Oct 04, 2020 at 11:16:10PM +0200, Ard Biesheuvel wrote:
> > On Sun, 4 Oct 2020 at 20:48, Stephan M??ller <smueller@chronox.de> wrote:
> > >
> > > The RISC-V architecture is about to implement the callback
> > > random_get_entropy with a function that is not exported to modules.
> >
> > Why is that? Wouldn't it be better to export the symbol instead?
>
> get_cycles is a low-level time keeping detail that really should not
> be exported, and at least for RISC-V this would be the only modular
> user.  Once that is sorted out I'll audit other common architectures
> to drop the export, as it isn't something that should be used in ramdom
> driver code.

Fair enough.

But this means we should fix the jitterentropy driver rather than
sidestepping the issue by only allowing it to be built in a way where
we don't happen to notice that the symbol in question is not meant for
general consumption.

If jitterentropy is a special case, we could put a alternate
non-'static inline' version of random_get_entropy() in the core
kernel, and only export it if JITTER_ENTROPY is built as a module in
the first place. But I'd prefer it if jitterentropy switches to an API
that is suitable for driver consumption.