Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Sun, 31 Jul 2022 02:23:49 +0200
From:   "Jason A. Donenfeld" <Jason@zx2c4.com>
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     linux-kernel@vger.kernel.org, linux-crypto@vger.kernel.org,
        x86@kernel.org, Nadia Heninger <nadiah@cs.ucsd.edu>,
        Thomas Ristenpart <ristenpart@cornell.edu>,
        Theodore Ts'o <tytso@mit.edu>,
        Vincenzo Frascino <vincenzo.frascino@arm.com>,
        Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>,
        Florian Weimer <fweimer@redhat.com>
Subject: Re: [PATCH RFC v1] random: implement getrandom() in vDSO
Message-ID: <YuXLlUZ8EzvZB43U@zx2c4.com>
References: <20220729145525.1729066-1-Jason@zx2c4.com>
 <CAHk-=wiLwz=9h9LD1-_yb1+T+u59a2EjTmMvCiGj4A-ZsPN1wA@mail.gmail.com>
 <YuXCpyULk6jFgGV5@zx2c4.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <YuXCpyULk6jFgGV5@zx2c4.com>
Precedence: bulk

On Sun, Jul 31, 2022 at 01:45:43AM +0200, Jason A. Donenfeld wrote:
> So, anyway, if I do muster a v2 of this (perhaps just to see the idea
> through), the API might split in two to something like:
> 
>   void *getrandom_allocate_states([inout] size_t *number_of_states, [out] size_t *length_per_state);
>   ssize_t getrandom(void *state, void *buffer, size_t len, unsigned long flags);
> 
> User code will call getrandom_allocate_state(), which will allocate
> enough pages to hold *number_of_states, and return the size of each one
> in length_per_state and the number actually allocated back in
> number_of_states. The result can then be sliced up by that size, and
> passed to getrandom(). So glibc or whatever would presumably allocate
> one per thread, and handle any reentrancy/locking around it.
> 
> Or some other variation on that. I'm sure you hate those function
> signatures. Everybody loves to bikeshed APIs, right? There's plenty to
> be tweaked. But that's anyhow about where my thinking is for a potential
> v2.

Doing this also doubled performance, perhaps unsurprisingly, as that
getcpu() operation wasn't free.

For uint32_t generation:

   vdso: 25000000 times in 0.289876265 seconds
syscall: 25000000 times in 4.296636025 seconds