Re: [PATCH] hw_random: rockchip: import driver from vendor tree

[Mikhail Rudenko]

On 2022-09-29 at 09:45 +02, LABBE Corentin <clabbe@xxxxxxxxxxxx> wrote:
> Le Tue, Sep 27, 2022 at 11:39:17PM +0200, Aurelien Jarno a écrit :
>> On 2022-09-27 21:46, LABBE Corentin wrote:
>> > Le Tue, Sep 27, 2022 at 06:37:53PM +0200, Aurelien Jarno a écrit :
>> > > On 2022-09-20 10:35, Robin Murphy wrote:
>> > > > On 2022-09-19 22:00, Jason A. Donenfeld wrote:
>> > > > > The Rockchip driver has long existed out of tree, but not upstream.
>> > > > > There is support for it upstream in u-boot, but not in Linux proper.
>> > > > > This commit imports the GPLv2 driver written by Lin Jinhan, together
>> > > > > with the DTS and config blobs from Wevsty.
>> > > >
>> > > > Note that Corentin has a series enabling the full crypto driver for
>> > > > RK3328 and RK3399[1], so it would seem more sensible to add TRNG support
>> > > > to that. Having confliciting compatibles for the same hardware that
>> > > > force the user to change their DT to choose one functionality or the
>> > > > other isn't good (plus there's also no binding for this one).
>> > >
>> > > It might make sense for the cryptov1-rng driver (I haven't checked). For
>> > > the cryptov2-rng driver, I looked at the RK3568 TRM (I can't find the
>> > > RK3588 one), and from what I understand crypto and TRNG are two
>> > > different devices, using different address spaces, clock, reset and
>> > > interrupts. The vendor kernel uses two different drivers.
>> > >
>> >
>> > I confirm that TRNG is not on the same IP on rk3568, something I didnt remark when doing my V2 driver. (I need to remove rng clock from rk3568 dt).
>> > But the rk3588 crypto IP and the TRNG are in the same device.
>>
>> Ok, thanks for confirming about the rk3568. It seems the only one in the
>> family with separate devices for TRNG and crypto. Does it means we need
>> a separate TRNG driver only for it? Or could we handle it the same way
>> than for instance rk3588 anyway?
>
> I just got now the part 1 of rk3588 TRM which I has missing and it show some conflicting information.
> rk3588 seems to have both a dedicated TRNG (TRNG_NS/TRNG_S) with dedicated address space and a TRNG inside the crypto IP.
> But for the moment, the TRNG inside crypto IP seems defective.

So what's the ultimate decision? Does anyone work on merging this into
the existing crypto driver? I have a use case with an rk3399-based board,
where having hardware rng enhances boot times dramatically (at least for
some userspaces; see also [1]).

[1] https://bugzilla.kernel.org/show_bug.cgi?id=216502

--
Best regards,
Mikhail Rudenko