A20 hardware crypto

[disconnect]

After i've upgraded to 5.00, and upgraded the legacy kernel to vanilla (mostly for hardware crypto), i've noticed strange errors: blocky video play from dlna, and from samba too. So i did some test, copied few files on my cubieboard2, and compared the sha hashes of both version, and sometimes the cb2 version is changed, sometimes not.

 

Also i see that a lot of synced files became different on cb2.

 

Is there a way to do some in memory tests to verify how the hw crpyto works? For example random numbers -> hw encrypt -> decrypt -> check if still the same.

 

These test i've made:

-as i'm using the encrypted HDD as a backup storage too, i've compared the source files with the backup files, and the most of them was different

-created an 1gb file with zeroes on sd card, then copy to the encrypted HDD (connected with sata)

-sdiff/diff tells the files are differ after the copy.

-copied multiple files on the encypted hdd (from pc to cubie), and did some sha1 checks. The <1GB files were mosly identical, but the >1GB files always had difference.

-disconnected the encrypted HDD, and connected to my main pc, and checked these files again, and all were indentical, including the backup files, and all were identical as should be

 

So it seems the hardware encryption works without errors, but the hw decryption is unreliable, some parts of files are decrypted wrongly.

The HDD is encrypted with LUKS aes-cbc-plain64 sha1, ext4 fs.

[disconnect]

Ok i figured out what is rmmod used for.

 

So if sun4i_ss is loaded, then the backup files are different

If that module is removed, then everything is OK (except of much slower transfer rates, and high cpu usage).

I wonder if this problem is connected to the slow hashing problem.

[disconnect]

I've tried it with a fresh jessie install, same happens: software deccryption OK, but if i load sun4i_ss decryption goes wrong.

 

Perhaps the problem with sun4i_ss, or just a misconfiguration?

[Igor]

I think you should report this here:
 

http://sunxi.montjoie.ovh/

and / or

https://groups.google.com/forum/#!forum/linux-sunxi

[disconnect]

I've reported. The developer of sun4i-ss says i should try kernel 4.4.2, so i read the changelog, and there are a lot of fixes for crypto and af_alg indeed. But i don't want to compile kernel. When armbian will be updated to kernel 4.4.2 or better?

[zador.blood.stained]

Please try kernel from this topic (again?  ). I updated it to 4.4.2 yesterday.

[disconnect]

Thanks, i've tried, but still the same

When the next update comes through apt-get upgrade, will it overwrite this? Or is should restore the "official" next kernel?

[zador.blood.stained]

When the next update comes through apt-get upgrade, will it overwrite this? Or is should restore the "official" next kernel?

It will overwrite this, so you can leave this kernel as it is for now.

[disconnect]

This is interesting: https://www.mail-archive.com/linux-crypto@vger.kernel.org/msg17926.html

[zador.blood.stained]

This is interesting too (starting with 18:25 timestamp): http://irclog.whitequark.org/linux-sunxi/2016-03-11

I think this is about mainline kernel

[disconnect]

Patch (workaround) for the problem: https://lkml.org/lkml/2016/3/23/302