meeting 2023-12-15 #12

Merged
neil merged 1 commits from meeting/2023-12-15 into main 2023-12-15 17:05:25 +00:00

View File

@ -0,0 +1,103 @@
# SIG/AltArch meeting 2023-12-15
## Attendees:
* Sherif
* Bryan
* Pablo Greco
* Neil Hanlon
* Skip Grube
## Follow Ups
* Libre Compute - Try/Investigate using vendor tree to build uboot - Pratham
* Investigate `arm-image-installer` script/tool for helping write disk images for specific boards - Sherif (?)
* Odroid N2 - Track installability
* Orange Pi 5 - Track installability
* Rock5b - Track installability
* Libre Computing Boards (3)
* Maintenance - Importing dependencies from Fedora/EPEL
* Neil is working on this as part of SIG/HPC, too
* Document process on backporting uboot and kernels - Pablo
* To include information on how we work and interact with upstreams, with examples! - Sherif
* Check and investigate which boards have hardcoded grubx64 paths in their uboot; address them
## Discussions
* Neil provided Sherif with 2 RPi CM4s (+ IO boards), an RPi5, and a SiFive VF2 (RiscV) at SC23 in November
* Need to work on integrating SBC and related changes into SIG/Kernel mainline in order to have unified kernel
* Bryan has tried mainline kernel, but it isn't booting -- ends up in dracut shell.
* Will figure out how to capture logs via serial
* Libre Tritium H5
* It's all Skip's fault
* 32-bit arm (armhfp/arm7vl) Rocky Port
* Sherif has some odroid boards from work he can donate
* Odroid XU4 x2
* Odroid XU4Q x1 - XU4 w/ massive heatsink
* Current status (per Pablo)
* have the full bootstrap up to 9.1; we need to start building atop it
* need to add Peridot support for armhfp (i can't remember if we included the arch as an option)
* where will we host the odroid boards from Sherif
* Most of the SBCs for arm7vl/armhfp are not powerful enough to be good build hosts. Maybe we should try Honeycombs? (Solid-run)
* Pablo used his honeycomb and a fedora image to do the initial bootstrap builds
* Neil will reach out to OSU about availability of Lenovo eMAGs for us
* has neil spoken to Fabian? - No ; Neil will do this
* RISCv Rocky Port Status
* Neil - Investigate what and how much riscv hardware we need.. and where we'll host it
* This probably is going to be a whole bunch of VF2s
* P670 (16-core - MilkV oasis) would be great, but, upstream support is not yet there. VF2 is a better target for Rocky 9
* Various conversations happened here but Neil was dealing with keeping all the mice in his house dead. The following is second-hand :)
* Rpi Kernel - Should it live?
* We want to get rid of it, but we can't yet. Skip says yes
* Plan:
* Build the rpi kernel as a variant of the mainline
* rpi kernel will be subsumed into kernel-mainline-sbc, not further upstream
* This will allow more obvious contribution to ELrepo for SBCs
* RPI image will be replaced with a new GenericArm-SBC image produced by Empanadas which includes kernel-mainline-sbc
* Next meetings
* Dec 29 - Cancel.
* Jan 12 - Keep
## Action Items
* Neil to actually make tickets for the follow ups / decisions. But, like, for real this time.
* Neil to reach out to people re: ARMHFP hardwares
* Ampere
* OSUOSL
* Fabian (arrfab)
* Neil to figure out what RISCv hardware we need
* Announce Dec 29th meeting cancelation
## Old business
### 2023-10-20
#### Action Items
* Libre Compute - Try/Investigate using vendor tree to build uboot - Pratham
* Investigate `arm-image-installer` script/tool for helping write disk images for specific boards - Sherif (?)
* Odroid N2 - Track installability
* Orange Pi 5 - Track installability
* Rock5b - Track installability
* Libre Computing Boards (3)
* Maintenance - Importing dependencies from Fedora/EPEL
* Document process on backporting uboot and kernels - Pablo
* To include information on how we work and interact with upstreams, with examples! - Sherif
* Check and investigate which boards have hardcoded grubx64 paths in their uboot; address them
* Acquire armhfp (arm7vl) hardware - Neil to reach out to OSU OSL
* Neil - Investigate what and how much riscv hardware we need.. and where we'll host it
#### Decisions
* Images should be generic so as to be able to be installed on any board, where possible.
* uboot can live on SPI or external flash
* We need to help document this for our users and have guides on how to get the SBCs working. This is a big pain point for many SBC users as there is so much variation.
* To this end, we should investigate some tooling to aid in writing images for boards which will boot on the first try.
* Uboot should similarly be generic and work for all boards, insofar as it is practical to do so.
* Some boards this doesn't make sense for, e.g., ones that have uboot from 2014 + hundreds of patches, however, these are few and far between. For these, we can package them as RPMs and provide them (see note below re: licensing)
* We will wait for the next Kernel.org LTS to be cut, as that should have all the changes for the boards we're talking about.
* We will engage with our upstreams (Fedora, ELRepo, etc) for changes we make with the Kernel SIG for AltArch/SBC support.
* This necessitates participating in SIG/Kernel to represent our needs in these kernels.
* We should strive to perform native builds when possible, but recognize that emulation is a necessary evil.
!!! note "uboot and other licensing"
We need to ensure that we are careful about the licensing of softwares we wish to include in the SIG distribution.