Re: [PATCH dunfell] cryptodev: Move to 1.12 revision

[Nishanth Menon <nm@ti.com>]

On 19:58-20210304, Denys Dmytriyenko wrote:
> On Thu, Mar 04, 2021 at 05:14:34PM -0600, Nishanth Menon wrote:
> > On 14:48-20210304, Denys Dmytriyenko wrote:
> > > BTW, you'd need to ensure updating cryptodev from 1.10 to 1.12 does not affect 
> > > current 5.4-based builds and/or releases. I'm sure it will build fine against 
> > > this older kernel (most of Makefile changes), but concerned about any run-time 
> > > regressions (the actual code changes).
> > 
> > Thank you, actually you bring out a very good point here. While I know
> > we can create our own internal private fork of arago for upstream
> > component testing, I am starting to wonder if creating either of:
> > 
> > a) meta-ti-mainline that builds on top of meta-ti
> > OR
> > b) meta-arago-mainline on top of meta-arago
> > 
> > is a smarter approach - personally, I prefer (a)? While, I don't want
> > to end up creating too many layers, but your point is valid that I
> > should also be careful to not mess with folks using the meta-arago in
> > production environments having to deal with challenges we are trying
> > to flush out by testing the bleeding edge of kernel - and I'd like to
> > make sure TI ecosystem is able to leverage/contribute as well (an
> > internal fork will not be that useful)..
> 
> First of all, an internal fork was an unintended consequence. But since we are 
> discussing this in a public forum, I won't go into more details. :)

:) understood :)

> 
> Creating yet another layer is certainly an option. For mainline purposes (a) 
> is definitely a better choice. And I assume that would be public...

yes, that is my thought.

> 
> On the other hand, if you only need this for couple of recipes, you can do 
> this with alternative providers. E.g. instead of altering existing recipes 
> with bbappends (e.g. cryptodev), you can have many providers for the same 
> package (e.g. linux-ti-staging, linux-ti-mainline, even linux-yocto from 
> OE-core all provide the Linux kernel, or virtual/kernel; same with u-boot 
> from OE-core and u-boot-ti-staging plus u-boot-ti-mainline provide U-boot 
> bootloader, like virtual/bootloaader). Those can be switched with a global 
> PREFERRED_PROVIDER_<pkg> variable:
> 
> https://git.yoctoproject.org/cgit/cgit.cgi/meta-ti/tree/conf/machine/include/k3.inc#n13
> 
> In other words, if cryptodev is the only such case right now, it's less 
> involved to create cryptodev-ti-mainline recipes, set those to corresponding 
> PROVIDES, as well as PREFERRED_PROVIDER_ in necessary config files. Let me 
> know if you need help with that.


Hmm.. Thanks.. yeah, it is cryptodev today, something else tomorrow,
in addition, I need to expand from purely kernel and u-boot upstream
components alone to TF-A, linux-firmware upstream and eventually few
additional packages etc. Mostly as a forward looking layer to uncover
issues ahead of time. My thoughts on this is as follows: While the
provider view can easily fit as well, it is a little less distracting
or in some cases, un-intended usage to keep a "experimental" and
"forward looking" layer away from users of production layer - then the
distinction is very clearcut and risks(essentially "you are on your
own") associated with the same would be clear as well. Who knows, some
day, we might be even be able to delete such a layer since upstream will
be rocksolid (touch wood)..


-- 
Regards,
Nishanth Menon
Key (0xDDB5849D1736249D) / Fingerprint: F8A2 8693 54EB 8232 17A3  1A34 DDB5 849D 1736 249D