AES-CFB-128 test failed (using wpa_supplicant)

[Jussi Kivilinna]

Hello,

On 07.06.2017 00:13, Belisko Marek wrote:
> Hi Jussi,
> 
> On Tue, Jun 6, 2017 at 6:55 PM, Jussi Kivilinna <jussi.kivilinna at iki.fi> wrote:
>> Hello,
>>
>> On 06.06.2017 10:17, Belisko Marek wrote:
>>
>> Looks like there is wrong use of assembly operand constraints in Padlock
>> accelerated AES code. Can you test if the attached patch helps? Patch is
>> for libgcrypt-1.6.x.
> After applying patch there are no more erro messages in syslog and I
> can connect to wifi successfully.
> Thanks a lot. Is this patch present in newer version of libgcrypt? Thanks.

Not yet, I'll prepare patch for development branch later this week.

The problem is less likely to manifest in 1.7 and later since
the Padlock assembly function is no longer inlined to bulk AES
functions (in this case CFB). This is because as Padlock function
has moved to separate sourcefile and behind indirect function
call, so compiler is much less likely to generate code that
break.

-Jussi