Re: [Ocf-linux-users] crypto_newsession / crypto_freesession

[David M. <Dav...@se...>]

Jivin Pirasenna Thiyagarajan lays it down ...
> I'm writing a driver and attempting it to make it OCF compatible.  I was
> intending on passing the virtual memory address I allocate for driver data
> structures for a given operation as the session id.  To avoid 32/64bit
> issues, having the crypto_newsession pass in u_int64_t * would make it
> convenient.  Otherwise, I'll have to keep a mapping table and look up
> routines appropriately (unnecessary computation cycles).

Hmm,  sounds fair,  only I think that crypto.c messes with the 64 bit
version a fair bit before handing it to the driver.

Check crypto.c:crypto_newsession to see this in action.

The 64bit version has the driver id + the drivers session id and the
drivers capability flags all merged into it.

It's probably unlikely your card with have more than 4GB of memory, or
maybe it does.  You could just return the "offset" into it's memory for
that session.  Saves a lookup table (baseaddr + offset = address).

> From the looks of it, it seems just like an interface definition issue,
> since as I pointed out, from within the ioctl, it does use u_int64_t
> internally.

Unfortunately I think it's not that simple, happy to consider something
though if you have ideas.

Are you planning to do a FreeBSD driver as well ?  If so it wouldn't hurt to
check if it's still the same.  Otherwise I'll get around to checking in
time ;-)

Cheers,
Davidm


> On Wed, Jan 28, 2009 at 3:58 PM, David McCullough <
> Dav...@se...> wrote:
> 
> >
> > Jivin Pirasenna Thiyagarajan lays it down ...
> > > I've used OCF patch and attempting to use the cryptodev interface.  In
> > > ocf-compat.h, I see crypto_newsession's session ID is passed in as a
> > > u_int32_t *, whereas, the crypto_freesession takes the session id as
> > > u_int64_t.  It seems like this is in error.  Looks like sid data type
> > should
> > > be u_int64_t *sidp.  In cryptodev.c where it is used, the sid is defined
> > as
> > > u_int64_t in cryptodev_ioctl.
> >
> > I had to look twice at that too now that you point it out.
> >
> > The trick is that device_method_t structure and resulting
> > CRYPTODEV_NEWSESSION and cryptodev_newsession used for it are
> > part of the interface for OCF drivers.  This is how OCF communicates
> > with it's crypto drivers (ie., safe, hifn7751, cryptosoft etc).
> >
> > cryptodev,  the user interface driver,  doesn't talk to the crypto
> > drivers,  it uses the OCF API's:
> >
> >        crypto_newsession
> >        crypto_freesession
> >
> > and they are both u_int64_t.
> >
> > Are you having a problem with cryptodev or did you just notice this oddity
> > in the code ?
> >
> > I admit the naming is confusing,  but we inherited that with that last
> > round of BSD updates.  Unless you are writing a driver,  ignore the section
> > at the top of ocf-compat.h,  in fact,  ignore as much of ocf-compat.h
> > as you can every chance you get :-)  It's where we hide the older OS
> > support and some BSD compat things to try and keep the driver source as
> > close to the original as practical.
> >
> > Cheers,
> > Davidm
> >
> > --
> > David McCullough,  dav...@se...,   Ph:+61
> > 734352815
> > Secure Computing - SnapGear  http://www.uCdot.org
> > http://www.snapgear.com
> >
> 
> 
> 
> -- 
> Online Gallery: http://www.deptons.com
> You comments and ratings are very welcome!!

-- 
David McCullough,  dav...@se...,   Ph:+61 734352815
Secure Computing - SnapGear  http://www.uCdot.org   http://www.snapgear.com