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;
MIME-Version: 1.0
References: <CAJKOXPfVXnazraHq-NCCM+h26FFi1U6Z3OSYOX17XH3bj-UDLQ@mail.gmail.com>
 <3053033.se6UkFii4W@tauon.chronox.de> <CAJKOXPd7JdzM2_PiHkS1ix1yU9hZmMbGDZaL07vbON0XzORH=w@mail.gmail.com>
 <2002208.sm9qV7UXju@tauon.chronox.de>
In-Reply-To: <2002208.sm9qV7UXju@tauon.chronox.de>
From:   Krzysztof Kozlowski <krzk@kernel.org>
Date:   Thu, 30 Aug 2018 15:54:19 +0200
Message-ID: <CAJKOXPdh6QReKb+Lt-_xTpRuzFbO-O-GVg+T1aA-9ST94oimwg@mail.gmail.com>
Subject: Re: Locking for HW crypto accelerators
To:     smueller@chronox.de
Cc:     herbert@gondor.apana.org.au, davem@davemloft.net,
        linux-crypto@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 30 Aug 2018 at 15:19, Stephan Mueller <smueller@chronox.de> wrote:
>
> Am Donnerstag, 30. August 2018, 15:09:05 CEST schrieb Krzysztof Kozlowski:
>
> Hi Krzysztof,
>
> > Thanks Stephan for hints. Let's assume the each of init, update and
> > final are atomic... but how about the relation between update and
> > final? I have two concurrent users in user-space but only one HW:
> >
> > Process A:             Process B:
> > init() and set_key()
> >                        init() and different key
> > update(some_data)
> >                        update(different_data)
> > final()
> >                        final()
> >
> > The final() from process A will now produce the result of hashing/CRC
> > of some_data and different_data (and even maybe mixed with init() for
> > different key). All because in the meantime process B added its own
> > data to the HW.
>
> The question is where is the state of the cipher operation kept that is
> produced with each of the init/update/final calls. Your answer implies that
> this state is kept in hardware.

Yes, that's correct. The HW once initialized to specific CRC
parameters (size, polynomial, initial value), will operate on this
till another init happens.

> But commonly the state is kept in software. Look at ahash_request for example,
> the __ctx pointer is intended to point to the memory the driver needs to store
> its state.
>
> Pick a random driver implementation and search for ahash_request_ctx, for
> example.

I see now some drivers are indeed saving and restoring state.

Thanks!

Best regards,
Krzysztof