aesni(?) corrupts data on 8.2-BETA1

[Jan Henrik Sylvester]

On 12/12/2010 15:26, Mike Tancsa wrote:
> On 12/12/2010 3:43 AM, Kostik Belousov wrote:
>> Please try this patch on the latest HEAD or RELENG_8.
>>
>
> I tried both on i386 and amd64 and all looks good!  I did 1000
> iterations of
> cryptotest -c -z -t 10
>
> without issue!  Thanks for the quick fix!

I redid my testing on a (read-only attached) geli provider (now with XTS 
instead of CBC) taking checksums of all files:

8.2-BETA1 without patch without aesni loaded: all checksums ok.
8.2-BETA1 with patch without aesni loaded: all checksums ok.
8.2-BETA1 without patch with aesni loaded: 124 of 18209 failures.
8.2-BETA1 with patch with aesni loaded: all checksums ok.

The patch seems to fix my geli problem on 8.2-BETA1 amd64, too.

Thanks!
Jan Henrik