On Wed, Dec 9, 2009 at 4:07 AM, Doug Shupe <[email protected]> wrote:
> Steve,
> You should read up on the HMC configuation of the CRYPTO processors. Maybe a
> Redbook too.
> As I recall, (old memories), they are not able to be brought online until
> they have been configured though I could be mistaken.
> VM directory statements of interest: for LPAR LPA7.

You're talking about the PCI Crypto cards. Those are different beasts
(that came in at least 4 different shapes and behave like a chpid with
I/O devices). They RS6K (or whatever is on those cards) provides some
statistics in the channel measurement blocks (varying per type of
card).

Steve was talking about the cryptic instructions implemented in the
CPU itself. The application will have to determine whether those
instructions are available (either through the STFLE that Alan showed,
or by just testing the ice). If the hardware support is not there (or
the conversation chose a symmetric cipher that is not supported by the
hardware) the application will have to perform the algorithm with more
traditional instructions (that take more time).

That logic is in the "libica" libraries of the OpenCryptoki project.
Configuration is not entirely trivial, so indeed wise to double check
that you really are using it. Since the CPACF instructions are
performed under SIE, CP has no knowledge whether your server uses them
or not. The book l26xci00.pdf refers to the "icastats" command in
openCryptoki v2 to display such statistics. My distribution comes with
v1 and the "icainfo" only shows which ones are available, not whether
an application takes advantage of them or maybe uses another library
that only does software emulation.

PS For those unaware (like myself) who think they can measure the
effect by looking at the CP usage for the guest, be aware that the
crypto drivers have an option to busy-wait for the crypto engine, so
the difference may be less obvious (unless you also have the Linux
process data and see excessive CPU usage in the associate daemon).

Rob
-- 
Rob van der Heij
Velocity Software
http://www.velocitysoftware.com/

Reply via email to