On Sat, Aug 06, 2016 at 01:22:01AM +0200, Michael Weiser wrote:
> Hi Phil,
> On Fri, Aug 05, 2016 at 10:09:08PM +0200, Phil Sutter wrote:
> > > I've done some groundwork, mostly research through a well known search
> > > engine and in the kernel code. What I've found is now at
> > > https://github.com/michaelweiser/cryptodev-linux/commit/30e66ed072d34ad0ca93f7457a9772519ca68de0.
> > Looks pretty reasonable. The only thing I would worry about is the
> > dropped checks in lines 210 and 228. Was there a specific reason for
> > that?
> Yeah, they check for NULL pointers themselves. I've seen checks like
> those dropped in Herbert Xu's patches and looked at the code of
> skcipher_request_free() to find kzfree() which in turn contains:
>         if (unlikely(ZERO_OR_NULL_PTR(mem)))
>                 return;
> See https://www.spinics.net/lists/linux-crypto/msg18154.html line 120.

Ah, I see.

> > > It compiles but I haven't had a chance to actually test it for function
> > > yet. 
> > Well, after compiling and loading the module, you could run the tests in
> > test/ subdir (note how I subtly try to push you into testing).
> I expected nothing else and you've done it very smoothly. ;) I was on
> the train when I did that research and could only do some haphazard
> compiling of the module on my Mac. Now that I'm at home I've got a huge
> agenda, containing some vacation, for the next week, so I most likely
> will only get to actual testing by the week after next.

No problem, maybe I'll find time myself over the next few days.

> > > What is your take on supporting old kernel versions: Should I #ifdef for
> > > < 4.3 or can I just drop ablkcipher like I've done now?
> > Yes, sadly we have to support older kernels as well (see existing compat
> > code). That's the curse of out-of-tree modules, I fear.
> Are a ton of #ifdefs alright then or do you prefer another approach?

Well, doing it as beautiful as possible is preferred, but eventually it
will boil down to a ton of #ifdefs anyway. So no need to bust a gut over
it. :)

Thanks, Phil

Cryptodev-linux-devel mailing list

Reply via email to