Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 1 Apr 2020 21:11:04 +0000
From:   George Spelvin <lkml@SDF.ORG>
To:     linux-kernel@vger.kernel.org, tytso@mit.edu
Cc:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>, lkml@sdf.org
Subject: Re: [RFC PATCH v1 36/50] random: Merge batched entropy buffers
Message-ID: <20200401211104.GA2013@SDF.ORG>
References: <202003281643.02SGhLiv003379@sdf.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <202003281643.02SGhLiv003379@sdf.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

I just noticed a rather insidious bug in the preceding, so please
revoke my S-o-b.

Storing the batch position in the first byte works fine if it's
updated very late in get_random_uXX(), after the random value is
read from a refilled batch.  The first few versions of my code
did this.

But then I discovered the "xor trick" to handle unaligned reads
and it hugely simplfiied the code.  It simplified it so much that
the unaligned position wasn't needed much; only to compute the
aligned position.

While squashing together the various revisions to this code, I
moved the write-back of the position earlier in the code..  Which
violates the requirement stated in paragraph 2.  :-(

There are several possible fixes, but the simplest is to move
the "u8 position;" down a couple of lines, out of the union.
Since the following patch reduces the lock to a single byte,
there's room in the structure without increasing its size.

Revised patch will follow.