linux-kernel - Re: [PATCH v3 2/5] crypto: caam: Fix endless loop when RNG is already initialized

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Mon, 5 Feb 2018 13:54:43 +0000
From:   "Auer, Lukas" <lukas.auer@...ec.fraunhofer.de>
To:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "aymen.sghaier@....com" <aymen.sghaier@....com>,
        "horia.geanta@....com" <horia.geanta@....com>,
        "pure.logic@...us-software.ie" <pure.logic@...us-software.ie>,
        "linux-crypto@...r.kernel.org" <linux-crypto@...r.kernel.org>
CC:     "peng.fan@....com" <peng.fan@....com>,
        "davem@...emloft.net" <davem@...emloft.net>,
        "ryan.harkin@...aro.org" <ryan.harkin@...aro.org>,
        "fabio.estevam@....com" <fabio.estevam@....com>,
        "rui.silva@...aro.org" <rui.silva@...aro.org>,
        "herbert@...dor.apana.org.au" <herbert@...dor.apana.org.au>
Subject: Re: [PATCH v3 2/5] crypto: caam: Fix endless loop when RNG is
 already initialized

On Mon, 2018-02-05 at 08:45 +0000, Horia Geantă wrote:
> On 2/2/2018 2:54 PM, Auer, Lukas wrote:
> > On Fri, 2018-02-02 at 11:20 +0000, Bryan O'Donoghue wrote:
> > > On 01/02/18 12:16, Horia Geantă wrote:
> > > > If the loop cannot exit based on value of "ret" != -EAGAIN,
> > > > then it
> > > > means
> > > > caam_probe() will eventually fail due to ret == -EAGAIN:
> > > > 	if (ret) {
> > > > 		dev_err(dev, "failed to instantiate RNG");
> > > > 		goto caam_remove;
> > > > 	}
> > > 
> > > For me it's an endless loop applying the first two
> > > 
> > > https://patchwork.ozlabs.org/patch/866460/
> > > https://patchwork.ozlabs.org/patch/866462/
> > > 
> > > but not this one
> > > 
> > > https://patchwork.ozlabs.org/patch/865890/
> > > 
> 
> [snip]
> > 
> > I think the problem lies in the instantiate_rng() function. If the
> > driver is unable to acquire DEC0 it'll return -ENODEV. This should
> > terminate the while loop in the probe function. However, the return
> > value is never checked and is instead overwritten with -EAGAIN,
> > causing
> > the endless loop.
> > 
> > This problem only occurs if u-boot instantiates only one of the
> > state
> > handles (ent_delay doesn't get incremented) and the kernel runs in
> > non-
> > secure mode (DEC0 can't get acquired). Instantiating all state
> > handles
> > in u-boot therefore fixes this problem. In addition, the return
> > value
> > in instantiate_rng() should be handled correctly by including
> > 
> > if (ret)
> > 	break;
> > 
> > right after "ret = run_descriptor_deco0(ctrldev, desc, &status);".
> > 
> 
> Indeed, the error path is incorrect and should be fixed as you
> mentioned.
> I will send a patch replacing this one.
> Note that this fixes only the error path, meaning caam_probe() won't
> go into an
> endless loop and instead will return -ENODEV, due to being unable to
> acquire
> control of DECO0.
> 
> There are still a few hurdles to cross for CAAM to work in a TZ
> environment.
> 
> For e.g. could you please check / confirm whether DECO0MIDR (DECO0
> MID registers
> @0xA0, @0xA4) are set such that Linux kernel is allowed to r/w DECO0-
> related
> registers?
> 
> Thanks,
> Horia

On my board DECO0 MID ms is set to 0x8001, which I believe (going by
the structure of the other MID registers, since some of the bits are
only marked as reserved) is a MID of 1 (A7 cores) in secure mode.
Changing this to 0x9 for a MID of 1 in non-secure mode still fails the
DEC0 acquisition step in the probe call.

So unfortunately I am not sure what / if other steps are required to
use the CAAM in non-secure mode. Running a quick test with openssl
speed (using CAAM with cryptodev), it at least seems to be working.

Thanks,
Lukas