Contribute the A80 findings back to linux-sunxi.org #74

New issue

Open

opened 2026-08-30 14:26:49 +00:00 by tiagoagueda · 0 comments

tiagoagueda commented

2026-08-30 14:26:49 +00:00

Owner

Our A80 work has overtaken what linux-sunxi.org documents, in ways that are now provably wrong
on the wiki - and the wiki is where anyone else with this SoC will look first.

What the wiki currently says

A80: "Linux-sunxi support is not yet available, due to lack of devices and software."
Sunchip CX-A99: "Like all A80 based devices, there is no support in u-boot-sunxi. Like
all A80 based devices, there is no support in our sunxi kernel."
and, in its mainline support table, for DRAM: "Max 3.5 GiB / 2.0 GiB detected and enabled /
Kernel crashes with LPAE enabled which is required to access the last 0.5 GiB."

What we can demonstrate

That last row is the sharpest one. We run 4 GiB with LPAE enabled, soaked at 206.5 GiB
written and read back with zero mismatches (#61, 57-4gb-reclaimed.md), and the reason it works
is written down: mem= declares the high bank, and CMA has to be constrained below 4 GiB or
32-bit DMA masters cannot reach their buffers. The wiki's "kernel crashes with LPAE" is a real
observation with a now-known cause.

Beyond that we hold, with evidence: a booting mainline kernel and u-boot for an A80 board, the
real pinmap from this board's script.bin, DRAM parameters, thermal, USB3, GMAC, HDMI, S/PDIF,
an AR100 driver, and the OZ80120 A15 rail behaviour.

Why bother

Three reasons, in order of how much I believe them:

It protects the work. ARCHIVE.md is a long record of A80 material vanishing from every
mirror. Findings that exist only in one private repository are one disk away from the same
fate. The wiki is mirrored, archived and searched.
It is the cheapest form of upstreaming. Most of the tracker's upstream issues need
patches shaped for a mailing list. A wiki edit needs none of that, and it is where the next
person with an A80 starts.
It is a straightforward correction of statements that are now false.

Scope

Not "write everything up". Start with the highest-value, lowest-effort corrections:

the DRAM/LPAE row on Sunchip CX-A99, and the A80 memory-map consequence
a Tronsmart Draco AW80 Telos board page, following New Device Howto
the A80 support-status lines, which are simply out of date

Check the licence position first. linux-sunxi.org is CC BY-SA. Contributing our own text is
fine; anything derived from the vendor SDK or the datasheets is not ours to relicense, and the
A80 page already documents Allwinner's GPL posture for this SoC. Keep the two separate.

From a deep pass over the offline linux-sunxi.org mirror (a80/linux-sunxi.org/wiki-backup, 4095 pages, snapshot 2026-08-29). linux-sunxi.org content is CC BY-SA; the pages named above are the attribution.

Our A80 work has overtaken what linux-sunxi.org documents, in ways that are now provably wrong on the wiki - and the wiki is where anyone else with this SoC will look first. ## What the wiki currently says - **`A80`**: *"Linux-sunxi support is not yet available, due to lack of devices and software."* - **`Sunchip CX-A99`**: *"Like all A80 based devices, there is no support in u-boot-sunxi. Like all A80 based devices, there is no support in our sunxi kernel."* - and, in its mainline support table, for DRAM: **"Max 3.5 GiB / 2.0 GiB detected and enabled / Kernel crashes with LPAE enabled which is required to access the last 0.5 GiB."** ## What we can demonstrate That last row is the sharpest one. We run **4 GiB with LPAE enabled**, soaked at 206.5 GiB written and read back with zero mismatches (#61, `57-4gb-reclaimed.md`), and the reason it works is written down: `mem=` declares the high bank, and CMA has to be constrained below 4 GiB or 32-bit DMA masters cannot reach their buffers. The wiki's "kernel crashes with LPAE" is a real observation with a now-known cause. Beyond that we hold, with evidence: a booting mainline kernel and u-boot for an A80 board, the real pinmap from this board's `script.bin`, DRAM parameters, thermal, USB3, GMAC, HDMI, S/PDIF, an AR100 driver, and the `OZ80120` A15 rail behaviour. ## Why bother Three reasons, in order of how much I believe them: 1. **It protects the work.** `ARCHIVE.md` is a long record of A80 material vanishing from every mirror. Findings that exist only in one private repository are one disk away from the same fate. The wiki is mirrored, archived and searched. 2. **It is the cheapest form of upstreaming.** Most of the tracker's `upstream` issues need patches shaped for a mailing list. A wiki edit needs none of that, and it is where the next person with an A80 starts. 3. It is a straightforward correction of statements that are now false. ## Scope Not "write everything up". Start with the highest-value, lowest-effort corrections: - the DRAM/LPAE row on `Sunchip CX-A99`, and the A80 memory-map consequence - a `Tronsmart Draco AW80 Telos` board page, following `New Device Howto` - the A80 support-status lines, which are simply out of date **Check the licence position first.** linux-sunxi.org is CC BY-SA. Contributing our own text is fine; anything derived from the vendor SDK or the datasheets is not ours to relicense, and the `A80` page already documents Allwinner's GPL posture for this SoC. Keep the two separate. --- *From a deep pass over the offline linux-sunxi.org mirror (`a80/linux-sunxi.org/wiki-backup`, 4095 pages, snapshot 2026-08-29). linux-sunxi.org content is CC BY-SA; the pages named above are the attribution.*

tiagoagueda added the

infra

P3-normal

labels

2026-08-30 14:26:49 +00:00

tiagoagueda referenced this issue from a commit

2026-08-30 14:27:46 +00:00

Deep-read the linux-sunxi mirror against the tracker (#73 #74 #75)