marvell_cesa fails to register on kirkwood (88F6282)

[JM]

On Fri, Feb 5, 2016 at 9:41 PM, Arnaud Ebalard <arno at natisbad.org> wrote:
> Hi guys,
>
> Gregory CLEMENT <gregory.clement at free-electrons.com> writes:
>
>> Hi,
>>
>> +cc Arnaud
>>
>>  On ven., févr. 05 2016, Gregory CLEMENT <gregory.clement at free-electrons.com> wrote:
>>
>>> Hi JM,
>>>
>>>  On ven., févr. 05 2016, JM <fijam at archlinux.us> wrote:
>>>
>>>> On Fri, Feb 5, 2016 at 4:45 PM, Gregory CLEMENT
>>>> <gregory.clement at free-electrons.com> wrote:
>>>>> Hi JM,
>>>>>
>>>>>  On ven., févr. 05 2016, JM <fijam at archlinux.us> wrote:
>>>>>
>>>>>> On Fri, Feb 5, 2016 at 11:29 AM, Boris Brezillon
>>>>>> <boris.brezillon at free-electrons.com> wrote:
>>>>>>> Jan, Andrew
>>>>>>>
>>>>>>> On Fri, 5 Feb 2016 10:02:46 +0100
>>>>>>> Andrew Lunn <andrew at lunn.ch> wrote:
>>>>>>>
>>>>>>>> On Fri, Feb 05, 2016 at 06:01:18AM +0100, JM wrote:
>>>>>>>> > Hi,
>>>>>>>> >
>>>>>>>> > I am testing marvell_cesa on a kirkwood device (QNAP TS-212p, Marvell
>>>>>>>> > Kirkwood 88F6282, dts: http://pastebin.com/Gk95KFHf ).
>>>>>>>> > As of kernel 4.4 (4.4.0-trunk-kirkwood #1 Debian 4.4-1~exp1) it fails
>>>>>>>> > in the following way:
>>>>>>>> >
>>>>>>>> > modprobe marvell_cesa allhwsupport=1
>>>>>>>> > dmesg | tail -n 1
>>>>>>>> > [ 1057.855091] marvell-cesa: probe of f1030000.crypto failed with error -12
>>>>>>>> >
>>>>>>>> > or:
>>>>>>>> >
>>>>>>>> > modprobe marvell_cesa
>>>>>>>> > dmesg | tail -n 1
>>>>>>>> > [ 9178.399357] marvell-cesa: probe of f1030000.crypto failed with error -524
>>>>>>>> >
>>>>>>>> > With kernel 4.3 it would fail with error -22 on that hardware. There
>>>>>>>> > are scattered reports from other users experiencing the same issue
>>>>>>>> > (1,2,3)
>>>>>>>> >
>>>>>>>> > I wonder if this is a result of an invalid definition of the
>>>>>>>> > crypto-sram in the dts of this SoC, or perhaps some other problem
>>>>>>>> > (PEBKAC included).
>>>>>>>>
>>>>>>>> Hi Jan
>>>>>>>>
>>>>>>>> Do you have SRAM enabled in your kernel configuration. The
>>>>>>>> mvebu_v5_defconfig does not have it, which is a bug really.
>>>>>>>
>>>>>>> Hm, the CRYPTO_DEV_MARVELL_CESA option selects the SRAM one [1].
>>>>>>> Anyway, I also think this error is related to the SRAM (not sure why the
>>>>>>> driver is not able to get the SRAM though).
>>>>>>>
>>>>>>> Best Regards,
>>>>>>>
>>>>>>> Boris
>>>>>>>
>>>>>>> [1]http://lxr.free-electrons.com/source/drivers/crypto/Kconfig#L168
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Boris Brezillon, Free Electrons
>>>>>>> Embedded Linux and Kernel engineering
>>>>>>> http://free-electrons.com
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Thanks for getting back to me. Indeed, CONFIG_SRAM is enabled, I put
>>>>>> the complete .config here: http://pastebin.com/40eKMJqE
>>>>>>
>>>>>> If there is anything I can do to help squish this bug, please let me
>>>>>> know.
>>>>>
>>>>> How do you get your dts?
>>>>> it seems that you generated it from your dtb (or from /proc).
>>>>>
>>>>> I tried to do the same from arch/arm/boot/dts/kirkwood-ts219-6282.dtb,
>>>>> but the resulting dts is too different for an easy comparison.
>>>>>
>>>>> Could you also provide your dtb, maybe by using the same dtc with the
>>>>> option -s I will manage to get the something easier to compare.
>>>>
>>>
>>>> Yes, I simply used dtc -I fs /proc/device-tree to obtain the dts.
>>>>
>>>> Here is the relevant .dtb file : http://heap.ovh/files/kirkwood-ts219-6282.dtb
>>>>
>>>
>>> OK so this dtb is exacly the one from the mainline. As most of the CESU
>>> stuff is set in te dtis, we will be able to compare it on the other
>>> kirkwood based board.
>>>
>>> Gregory
>>
>> So with the help of Boris we checked the ranges propreties, the reg
>> property and the clock too. Everyting seems OK.
>>
>> Arnaud,
>>
>> as you are the on who tested the cesa on kirkwood, do you remember if
>> you needed to add specific Kconfig or to made some change in the dts?
>
> Here is the result I get on my ReadyNAS Duo v2 (88F6282, see below) w/ a
> stock 4.4 kernel I just cross-compiled w/ a gcc 4.9.2. The kernel has no
> modules and uses an appended DTB (u-boot on the platform is pretyy old):
>
> root at mood:~# dmesg | grep -i cesa
> [    3.306135] marvell-cesa f1030000.crypto: CESA device successfully registered
>
> root at mood:~# uname -a
> Linux mood 4.4.0.duov2 #1 Fri Feb 5 21:11:25 CET 2016 armv5tel GNU/Linux
>
> root at mood:~# cat /proc/cpuinfo
> processor       : 0
> model name      : Feroceon 88FR131 rev 1 (v5l)
> BogoMIPS        : 1594.16
> Features        : swp half thumb fastmult edsp
> CPU implementer : 0x56
> CPU architecture: 5TE
> CPU variant     : 0x2
> CPU part        : 0x131
> CPU revision    : 1
>
> Hardware        : Marvell Kirkwood (Flattened Device Tree)
> Revision        : 0000
> Serial          : 0000000000000000
>
> I'll try and do the same now w/ CESA support compiled as module to see
> what I get. I'll keep you posted.
>
> As a side note, a simple grep on the .config file you provided at
> http://pastebin.com/40eKMJqE makes me wonder why it has some
> grsec options residue. Did you test on a stock 4.4 kernel or on
> something w/ additional patches? If the latter, could you check what a
> stock kernel gives?
>
> Cheers,
>
> a+

Thanks for the reply. Your cpuinfo is exactly the same as mine. I am
using the kernel as shipped by Debian (4.4 from experimental). I was
directed from a Debian ARM mailing list to report to upstream, but
perhaps it is a distribution-specific bug. I will try to build a
vanilla kernel, although that will take me some time (no
cross-compilation environment at hand). I will keep you posted.

Best regards,
Jan