GPU: no viable path to mainline support #25

New issue

Closed

opened 2026-08-27 23:22:28 +00:00 by tiagoagueda · 0 comments

tiagoagueda commented

2026-08-27 23:22:28 +00:00

Owner

The A80 has a PowerVR Rogue G6230 (Series6, ~2013). Software rendering only.

Recorded purely so nobody spends an evening rediscovering this. Still wontfix — it is a
statement of the situation, not a task.

Reworded 2026-08-28

This issue previously said "there is no mainline support and there will not be any". That
was true when written and is no longer accurate as a blanket claim, so it is corrected here
rather than left standing.

Mainline does have an open PowerVR driver. drivers/gpu/drm/imagination/, written by
Imagination themselves with Collabora, plus a matching Mesa Vulkan driver. "PowerVR is never
open" stopped being true around 2023.

It still cannot help this chip, for three independent reasons:

Wrong generation. That driver targets newer Rogue-derived IP — the BXS/AXE families,
such as the AXE-1-16M in TI's AM62. The G6230 is Series6, a decade older and not a matter
of adding a device ID.
Firmware. The driver does not talk to the GPU directly; it loads per-GPU signed
firmware published by Imagination for supported parts. None exists for the G6230, and
there is no commercial reason for one to. This is the hard wall — even a perfect kernel
driver would have nothing to load.
The vendor blobs are unusable. What exists for the A80 is an Android DDK against kernel
3.4: closed userspace bound to an ancient kernel ABI, not shimmable onto mainline DRM/KMS
in any maintainable way.

Getting past all three would take either Imagination choosing to support decade-old silicon,
or a from-scratch reverse-engineering effort of the Panfrost/Nouveau kind, which nobody has
begun for this generation.

So the defensible claim is "no path exists and none is plausible" — not "never" as a
matter of principle. Same practical conclusion, reasoning that holds up.

It is moot for this board regardless

There is no display output at all on sun9i mainline — no HDMI encoder exists (#22), so
nothing reaches a screen. And for the headless appliance role in
41-production-plan.md, 3D acceleration is irrelevant. Even in the
best case the GPU would be the last thing that mattered, well after HDMI.

Why this was reworded rather than left alone

An overconfident "never" in a closed issue is the kind of claim a future reader takes at face
value without re-deriving. It is logged in AI-ASSISTANCE.md alongside the
other overstatements this project produced, and TODO.md carries the same correction.

The A80 has a **PowerVR Rogue G6230** (Series6, ~2013). Software rendering only. Recorded purely so nobody spends an evening rediscovering this. Still `wontfix` — it is a statement of the situation, not a task. --- ## Reworded 2026-08-28 This issue previously said *"there is no mainline support and there will not be any"*. That was true when written and is no longer accurate as a blanket claim, so it is corrected here rather than left standing. **Mainline does have an open PowerVR driver.** `drivers/gpu/drm/imagination/`, written by Imagination themselves with Collabora, plus a matching Mesa Vulkan driver. "PowerVR is never open" stopped being true around 2023. **It still cannot help this chip**, for three independent reasons: 1. **Wrong generation.** That driver targets newer Rogue-derived IP — the BXS/AXE families, such as the AXE-1-16M in TI's AM62. The G6230 is Series6, a decade older and not a matter of adding a device ID. 2. **Firmware.** The driver does not talk to the GPU directly; it loads **per-GPU signed firmware** published by Imagination for supported parts. None exists for the G6230, and there is no commercial reason for one to. This is the hard wall — even a perfect kernel driver would have nothing to load. 3. **The vendor blobs are unusable.** What exists for the A80 is an Android DDK against kernel 3.4: closed userspace bound to an ancient kernel ABI, not shimmable onto mainline DRM/KMS in any maintainable way. Getting past all three would take either Imagination choosing to support decade-old silicon, or a from-scratch reverse-engineering effort of the Panfrost/Nouveau kind, which nobody has begun for this generation. So the defensible claim is **"no path exists and none is plausible"** — not "never" as a matter of principle. Same practical conclusion, reasoning that holds up. ## It is moot for this board regardless There is **no display output at all** on sun9i mainline — no HDMI encoder exists (#22), so nothing reaches a screen. And for the headless appliance role in [41-production-plan.md](41-production-plan.md), 3D acceleration is irrelevant. Even in the best case the GPU would be the last thing that mattered, well after HDMI. ## Why this was reworded rather than left alone An overconfident "never" in a closed issue is the kind of claim a future reader takes at face value without re-deriving. It is logged in [AI-ASSISTANCE.md](AI-ASSISTANCE.md) alongside the other overstatements this project produced, and `TODO.md` carries the same correction.

tiagoagueda

2026-08-27 23:22:28 +00:00