Received-SPF: pass (google.com: best guess record for domain of linux-crypto-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200127165646.19806-1-andrew.smirnov@gmail.com>
 <20200127165646.19806-7-andrew.smirnov@gmail.com> <VI1PR0402MB34857006B42E8F0DAFECC514981B0@VI1PR0402MB3485.eurprd04.prod.outlook.com>
In-Reply-To: <VI1PR0402MB34857006B42E8F0DAFECC514981B0@VI1PR0402MB3485.eurprd04.prod.outlook.com>
From:   Andrey Smirnov <andrew.smirnov@gmail.com>
Date:   Mon, 24 Feb 2020 08:37:27 -0800
Message-ID: <CAHQ1cqHDotfJVUM9+Dk7=LyFxDhSWmud66HiZSzbZkKtn9R0Ug@mail.gmail.com>
Subject: Re: [PATCH v7 6/9] crypto: caam - check if RNG job failed
To:     Horia Geanta <horia.geanta@nxp.com>
Cc:     "linux-crypto@vger.kernel.org" <linux-crypto@vger.kernel.org>,
        Chris Healy <cphealy@gmail.com>,
        Lucas Stach <l.stach@pengutronix.de>,
        Herbert Xu <herbert@gondor.apana.org.au>,
        Iuliana Prodan <iuliana.prodan@nxp.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        dl-linux-imx <linux-imx@nxp.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-crypto-owner@vger.kernel.org
Precedence: bulk

On Wed, Feb 12, 2020 at 2:41 AM Horia Geanta <horia.geanta@nxp.com> wrote:
>
> On 1/27/2020 6:57 PM, Andrey Smirnov wrote:
> > @@ -60,12 +65,12 @@ static struct caam_rng_ctx *to_caam_rng_ctx(struct hwrng *r)
> >  static void caam_rng_done(struct device *jrdev, u32 *desc, u32 err,
> >                         void *context)
> >  {
> > -     struct completion *done = context;
> > +     struct caam_rng_job_ctx *jctx = context;
> >
> >       if (err)
> > -             caam_jr_strstatus(jrdev, err);
> > +             *jctx->err = caam_jr_strstatus(jrdev, err);
> >
> > -     complete(done);
> > +     complete(jctx->done);
> >  }
> >
> >  static u32 *caam_init_desc(u32 *desc, dma_addr_t dst_dma, int len)
> > @@ -89,6 +94,10 @@ static int caam_rng_read_one(struct device *jrdev,
> >  {
> >       dma_addr_t dst_dma;
> >       int err;
> > +     struct caam_rng_job_ctx jctx = {
> > +             .done = done,
> > +             .err  = &err,
> > +     };
> >
> >       len = min_t(int, len, CAAM_RNG_MAX_FIFO_STORE_SIZE);
> >
> > @@ -101,7 +110,7 @@ static int caam_rng_read_one(struct device *jrdev,
> >       init_completion(done);
> >       err = caam_jr_enqueue(jrdev,
> >                             caam_init_desc(desc, dst_dma, len),
> > -                           caam_rng_done, done);
> > +                           caam_rng_done, &jctx);
> AFAICT there's a race condition b/w caam_jr_enqueue() and caam_rng_done(),
> both writing to "err":
> caam_jr_enqueue()
>         -> JR interrupt -> caam_jr_interrupt() -> tasklet_schedule()...
>         -> spin_unlock_bh()
>         -> caam_jr_dequeue() -> caam_rng_done() -> write err
>         -> return 0 -> write err
>

Yes, I thought it didn't really matter for calling
wait_for_completion(done), but now that I think on it again, it can
return wrong result code from vcaam_rng_read_one(). Will fix in v8.

Thanks,
Andrey Smirnov