Hey debian-kernel, Karsten Merker <mer...@debian.org> (2017-06-11): > as we appear to have the same underlying problem in bugs 864536, > 864457 and 856111, I personally think that adding the modules > necessary for i2c-support in d-i is worth another upload before > r0, provided the current diagnosis in 864457 is correct and > handling the additional work due to this is doable for everybody > involved. > > If the current diagnosis in 864457 is correct, not providing i2c > modules AFAICS will not only break d-i completely on the > Firefly-RK3288 (bug 864536) but also the following usecases: > > - all hd-media and thereby all offline installs > - all installations to USB-connected harddisks > - all non-serial-console installations due to non-working > USB keyboard support > > on all systems that use the AXP20x series of powermanagement > controllers, which is a significant part of the armhf platforms > that we provide installer images for: > > - A10-OLinuXino-Lime > - A20-OLinuXino-Lime > - A20-OLinuXino-Lime2 > - A20-OLinuXino-MICRO > - A20-Olimex-SOM-EVB > - BananaPi > - BananaPro > - Cubieboard > - Cubieboard2 > - Cubietruck > - Lamobo_R1 > - orangepi_plus > - pcDuino > - pcDuino3 > > AIUI, the following changes to the kernel package would be > needed: > > - add an i2c-modules config for armhf which includes the generic > i2c-modules config plus the i2c-mv64xxx and i2c-rk3x modules > - add the axp20x_usb_power module to the armhf kernel-image config > to address the specifics of bug #856111 (see Ben Hutchings' notes > at https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=856111#54) > > As that effectively only makes the same modules that we already > install on all armhf systems also available inside the d-i > environment, the chances of causing a regression by this change > are rather low. Size constraints are AFAIK not a problem on > armhf (in contrast to armel), so the aforementioned changes > should be rather low-risk.
Based on Karsten's input, I'm now rather convinced that adding a udeb before r0 would be the best way to deal with it. Even if some other modules are missing inside, we still have a chance of fixing this in a point release by just adding a few .ko files to an already existing udeb, instead of introducing it during a point release. I've just checked with Niels, and this looks like a sane approach from a release team point of view. Do you know if you'll be able to perform a new linux upload from the stretch branch in a relatively near future? The sooner we get fixes, the more we can run tests, and the saner release people will be. ;) Please let me know if I can help. KiBi.
signature.asc
Description: Digital signature
