DevBoards Program Meeting April 18, 2024

44 views
Skip to first unread message

Jeff Scheel

unread,
Apr 18, 2024, 7:32:20 AMApr 18
to RISC-V Development Board Program
All,

We will be having our next meeting in a couple of hours, today, Thursday, April 18, 2024 at the following time:
  • 3:00 PM UTC,  
  • 5:00 PM Central Europe Summer Time,
  • 8:30 PM India Standard Time,
  • 11:00 PM China Standard Time
  • 11:00 AM Eastern U.S. Standard Time,  
  • 8:00 AM Pacific U.S. Standard Time.
Here's the meeting link (which can be copied to your calendar):

The agenda will be:
  • Board and Project updates
  • Program updates from RISC-V
  • What's cool, especially from Embedded World and other events?
  • Graphics discussion
  • Next program discussion
As always, any topics, updates, or information can be added to our agenda document, Dev Board Group Minutes or raised at the beginning of the meeting.
-Jeff

--
Jeff Scheel (he/him/his)
Director of Technical Programs, RISC-V International

Jeff Scheel

unread,
Apr 19, 2024, 12:33:29 PMApr 19
to RISC-V Development Board Program
All,

Welcome to our new and newer members today.  I hope you'll come back and join us again in the future.

We covered alot of ground today.  Thanks, especially to Carl for his insight into the GPU challenges.  It sounds like work exists but that it may not be as big of a hill to climb as I anticipated.  I love good news!!!

Per convention, the official minutes and chat history from the meeting are in our Meeting Document.  I'm attaching a copy of the minutes only here for your convenience.  Please question, correct, and extend the information in the document or to the email list as you deem appropriate.

Attendees: Jeff, Greg, Darren, Paul Sherman, Lance, Heinrich, Bo Gan, Carl, Emil, Yu Bo, Andreas, Daniel, Axel


Board Updates:

  • None


Program Updates:

  • Debian

 

Munin for debci riscv64 workers will be available soon.

  • Fedora


  • SUSE

    • Looking to add native workers to OBS ( https://build.opensuse.org/monitor )

    • Investigating Milk-V Pioneer.

      1. Reminder from Emil: Also note that the C930 cores in the Pioneer like the other "1st gen" T-Head cores (C910 on TH1520 and C906 on the D1) suffer from a broken floating point rounding bug. So if your package test suites are very thorough they might fail :/ 

    • Working with Sophgo on upstreaming their support.

  • Canonical

    • U-Boot Milk-V Mars support has gone upstream.  Patches for Milk-V Mars CM sent upstream.

    • Linux Upstream kernel Mars board support improving: patches for PCI and device-tree are under review.

  • R9

  • oreboot

    •  PoC for mixed AMP+SMP platform and boot flow design is WIP

  • Firmware

  • RISC-V

    • Kendryte boards submitted to vendor and expect to ship soon

    • Status requested on Pioneer shipments.  No update yet.

    • Huashan Pi projects acceptance/rejection underway.

    • Rocky Linux has received their VF2 boards.

    • LicheePi Cluster shipping to OSU.

    • Still building future plans.  Envision 2 boards – developer board (Raspberry Pi form factor) and server board (miniITX).  Strongest candidates at this time Banana Pi and Milk-V Oasis, but may change due to availability.


What’s cool?:


Miscellaneous discussions:

  1. How might we build a “program” in advance of having actual hardware?

    • Should we poll our Distro and “Core Projects” about their needs for pending hardware?  If so, what are key questions?

    • Is there value in building a virtual project for external program?  We could assume “Raspberry Pi” form factor.

  2. GPU Driver Support on RISC-V Platforms

    • How do we tackle this ecosystem?

    • Carl’s update: Bug with drivers helped alot (see above github.com/torvalds item in What’s Cool)

      • Ubuntu 23.10 on Unmatched with an old Radeon HD6450 card “worked”, but took 5+ min to boot with a lot of errors. Newer GPUs that required larger drivers did not work.

      • Testing the Linux 6.8 kernel on Arch worked perfectly with both the HD6450 and an RX460 card, haven’t had a chance to test the 6.8 kernel on a Ubuntu 23.10 userspace

    • Imagination (IMG) is what most RISC-V Boards are using

      • The cores used by every board we are tracking are using IMG’s “Rogue Architecture” which is brand new IP and they have committed to Open Source driver support. These are also Vulkan first, Vulkan native. So the expectation is that Mesa and thus Zink (module of Mesa) will be providing legacy OpenGL/GL-ES/GLX calls and “wrap” them via Vulkan to the hardware

      • Carl reached out to the devs, and learned a lot: https://gitlab.freedesktop.org/frankbinns/powervr/-/issues/7 

        • The development has moved from individual contributor’s repos on freedesktop.org to a unified one: https://gitlab.freedesktop.org/imagination

        • The current upstream drivers from Imagination are gated with an Aarch64 Kconfig  but they were fairly confident that all the code was portable and was receptive to enabling RV64 gating as well

        • I had also asked for support for a series of boards, here are the requests, the GPU blocks (as far as I can tell) and their status:

          • TH1520 (Lichee Pi4A) using BMX-4-64, partial support in place

          • JH7100 (VisionFive 2, Milk-V Mars, etc) using BXE-4-32, partial support but no firmware yet. Should be an easy lift

          • Spacemit K2 (BPI-F3) using BXE-2-32, partial support but no firmware. Should be an easy lift to support.

          • SG3260 (Oasis) using AXT-16-512, no support yet and will be a rather heavy lift since this is a completely different IP track.

          • This may not be a major problem in the short term, because the Oasis specifically has a wide aperture PCI-express implementation, so using an off the shelf GPU should “just work” (much like SiFive Unmatched)

      • Also working on GPU support for U-Boot which is quite old for several boards.

        • StarFive’s downstream U-Boot has support for enough of the Framebuffer on the JH7110 to give a splash screen. May be possible to leverage that for U-Boot’s framebuffer console

        • It may be possible to then leverage that code to get some form of framebuffer running on the TH1520 as well. Need to do lots of porting to get to that point (the TH1520 vendor U-Boot tree is even older than the JH7110)

        • My goal is something akin to Tow-Boot - which gives a much more user friendly experience. Even if the kernel DRM/KMS doesn’t work, this will at least make it easier for users to change operating systems due to UEFI support

        • Heinrich suggested that U-Boot enabling a console may allow Linux to access it, but I will need to investigate that: https://docs.kernel.org/fb/efifb.html I’m not sure if U-Boot implements the EFI UGA architecture

      • RISC-V GPU SIG

        • The SIG is working with various Universities to make a GPU core using exclusively RISC-V IP

        • During Summit 2023 there were discussions around how to support legacy APIs like GL, and we agreed that skipping that and going straight to Vulkan (in the same footsteps of IMG Rogue and Intel Xe) would make life much easier, and Mesa/Zink can be leveraged to provide legacy support.


-Jeff

--
Jeff Scheel (he/him/his)
Director of Technical Programs, RISC-V International

Reply all
Reply to author
Forward
0 new messages