Hello,
 (note that the main mailing list is cryptodev-linux-devel@gna.org)

What is the output you get from "make check" in the cryptodev
distribution? Could you provide a backtrace of the gnutls-cli segfault?

regards,
Nikos

PS. About the openssl speed, you need to use -evp mode to enable other
engines

On 02/17/2012 10:51 AM, Frank wrote:

> Hi,
> 
> First of all, thanks for cryptodev-linux, it's very useful.
> However, I've run into some odd behaviour that you might be able to shed some 
> light onto.

> These are the specs:

> - On Marvell Kirkwood hardware
> - Running Debian Wheezy (Linux kernel 3.2.0)
> - cryptodev-linux 1.0
> - (Debian version of) openssl 1.0.0g with cryptodev support (both tried 
> -DHAVE_CRYPTODEV -DUSE_CRYPTODEV_DIGESTS -DHASH_MAX_LEN=64 and only 
> -DHAVE_CRYPTODEV)
> - (Debian version of) gnutls 3.0.12 with cryptodev support
> 
> This is the observed behaviour:
> - openssl confirmed to use cryptodev interface (hardware accelerated when I 
> load mv_cesa.ko) when I run cipher benchmarks (e.g. openssl speed -evp 
> aes-128-cbc -engine af_alg -elapsed) or when aes-128-cbc is used with scp.
> - nginx can no longer serve https connections when cryptodev module is loaded 
> (lots of entries in the error log with ' [alert] 3287#0: worker process 3327 
> exited on signal 11'), regardless whether mv_cesa is loaded or not
> - gnutls-cli --benchmark-ciphers produces the following output when cryptodev 
> is loaded, regardless of whether mv_cesa is loaded or not (only the 
> throughput figures differ)
>       Checking AES-128-CBC with SHA1 (16kb payload)... Processed 29.46 MB in 
> 2.00 secs: 14.72 MB/sec
>       Checking AES-128-CBC with SHA256 (16kb payload)... Processed 22.81 MB 
> in 2.00 secs: 11.40 MB/sec
>       Checking AES-128-GCM (16kb payload)... Processed 14.58 MB in 2.00 secs: 
> 7.29 MB/sec
>       Checking SHA1 (16kb payload)... Segmentation fault
> Without cryptodev loaded, it will benchmark all ciphers just fine.
> 
> P.S. I can run 'openssl speed sha1 -elapsed' with cryptodev loaded, but it 
> never seems to use mv_cesa
> 
> 
> _______________________________________________
> Cryptodev-linux-commit mailing list
> cryptodev-linux-com...@gna.org
> https://mail.gna.org/listinfo/cryptodev-linux-commit
> 



_______________________________________________
Cryptodev-linux-devel mailing list
Cryptodev-linux-devel@gna.org
https://mail.gna.org/listinfo/cryptodev-linux-devel

Reply via email to