Skip to content
17 changes: 12 additions & 5 deletions device/dipper/BRINGUP-M1.md
Original file line number Diff line number Diff line change
@@ -1,14 +1,14 @@
<!-- SPDX-License-Identifier: MIT -->
# Xiaomi Mi 8 (dipper): RAM-only first light

**Status (2026-11-11):** the newest RAM boot is the 2026-11-11 one in [NOTES.md]: #505's rebuilt image `665fcbd682cd8e64b3ffec1ff7f4ab9ca33a26b9b47b11073be7a3c267abf2b1` on kernel/DTS #220's final head, with every UFS partition read-only. Before it, on 2026-11-10, the combined #503/#505 non-NFS image `540b5d31ea73a31ade9a8d8cb5152cf701dc3c2ad8256734790b0ef2db6dbdd0` booted by RAM on the Xiaomi Mi 8 with kernel/DTS PR #220 at `055a4019998f22cc0e2a8afef7650274a6ce0657`. Its initramfs root had all 85 UFS partition flags at `ro=1`. `/sys/block/sda/ro` was `0`, but `/dev/sda` was absent; `sdb`–`sdf` were `ro=1`. The guard confirmed this state. `qcom_wdt` and its feeder ran, simpledrm registered `fb0`, USB/SSH worked, and `/proc/iomem` reserved `0x85fc0000–0x8a8fffff`, covering the former TZ range. The pinned rescue SSH key matched, and `reboot-to` returned the phone to fastboot. Physical input events were not captured on this exact image.
**Status (2026-11-12):** the newest RAM boot used the Dipper FTS5 node enabled directly in a local kernel tree. `stmfts` registered as `event2`; the screen was not visually verified, and no physical touch was captured on that boot. The preceding corrected overlay boot captured contact, changing X/Y and release. All recorded UFS partition flags were read-only, `/dev/sda` was absent, and `0x85fc0000–0x8a8fffff` remained reserved. No UFS partition was written. Rebuildable kernel DTS and repository overlay sources are now proposed in kernel PR #249 and this PR; the test images predate those commits. Earlier boots are recorded in [NOTES.md].

A separate RAM NFS-root run used image `a2d435a7430e7e4c145c61cdb9866d16929731859d91c7105e6bfcaf9e3e9dd7` with the rebased #503 init. BusyBox `mount` switched to the test NFS root; all six disk flags and all 85 partition flags were `ro=1`. Systemd reached the multi-user target with `RuntimeWatchdogSec=60s`, and SSH remained available. `reboot-to bootloader` returned the phone to fastboot. The exported root was a generic Arch base; this did not test the Omarchy desktop or link-loss recovery. Full results are in [NOTES.md].
The FTS5 driver from [touch #219] is merged into linux-omarchy-mobile. The
Dipper board DTS from [DTS #220] is merged too (2026-11-11), and the EA8074
panel driver is under review in [panel #218].
The EA8074 panel test is recorded in [NOTES.md]. The FTS5 touchscreen
still needs a hardware test.
The EA8074 panel test and the corrected FTS5 touchscreen hardware test are
recorded in [NOTES.md]. Both used RAM boots; no UFS partition was written.
On 2026-11-09, a guarded RAM boot was used to capture all 84 GPT partitions
other than `userdata` (the phone has no `super`); source and compressed-image
checksums passed. The separate live userdata archive also passed its checksum
Expand Down Expand Up @@ -59,6 +59,12 @@ archive's checksum file. Do not boot if either backup is missing or fails.
without changing UFS, and the screen stays lit because simpledrm holds the
panel's VCI and VDDIO rails. If it goes dark about 30 seconds in, record
`dmesg | grep -i -E 'regulator|simple'` and stop.
Before booting, decompile the exact packed DTB on the laptop with
`dtc -I dtb -O dts` and verify it contains both
`removed-region@86200000` and `removed-region@89100000`. On the phone,
before touching the screen or doing any other test, verify
`/proc/iomem` marks the complete `0x85fc0000–0x8a8fffff` range reserved.
If it does not, stop immediately and run `reboot-to bootloader`.
Pass only when `uname -r`, the `boot.img` sha256, all partition `ro` values,
whole-disk `ro` values, and `ufs-ro` lines from `dmesg` are recorded.
`/dev/sda` must be absent or present with `/sys/block/sda/ro=1` in NFS mode;
Expand Down Expand Up @@ -117,8 +123,9 @@ repository; the SSH public key and generated host key stay in the local output.
change was not retained. The 2026-11-10 NFS test root reached systemd with
`RuntimeWatchdogSec=60s`; NFS link-loss recovery and the Omarchy desktop on
that root remain untested.
- Test the FTS5 touchscreen changes and GPU rendering on the phone before
marking them supported.
- FTS5 controller probe and physical contact/coordinate/release events passed
on 2026-11-12 (see [NOTES.md]); Hyprland input delivery and GPU rendering
remain untested on the phone.

From this rescue image, an SSH session can return to fastboot with
`reboot-to bootloader`. The installed ROM's `recovery` image was verified from
Expand Down
58 changes: 51 additions & 7 deletions device/dipper/NOTES.md
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,9 @@ the facts needed for review.

| Date | `boot.img` SHA-256 | Source, as recorded | `/dev/sda` and UFS partitions | Checks and gaps |
| --- | --- | --- | --- | --- |
| 2026-11-12 | `ac93a5d9f6787153cf620704f192c22da91cf2041789f16f4214111b8458d226` | Test kernel source: base `2a318a96ae9d4c87e061da1da13c05b11d305658` plus the FTS5 DTS change now proposed as kernel PR #249 commit `0d548847248512b1e8e1d40d66a85b81fe5ed471`; enabled directly for this test | `/dev/sda` absent (disk flag not applicable); `sdb`–`sdf` disk flags not read; all 85 partition flags `ro=1` | Full reserved range, kernel `7.3.0-rc6-omamobile-push1`, `stmfts` on `event2`, no Oops/BUG/panic/WARN. `qcom_wdt` and feeder state: not read. Exact touch X/Y extrema were not retained. No physical touch event captured; no UFS write. |
| 2026-11-12 | `1bf0ffd5302fff7fa75f49382567063d7e03bd52083baa81a88107a47dcbdbd2` | The booted test overlay itself defined the full touch node; its exact source was not retained. The later rebuild uses #249's disabled node plus [`device/dipper/overlays/fts5-touchscreen.dtso`](overlays/fts5-touchscreen.dtso), blob `f09381d59a110d02a8b8819f585e304efb1870a3`; both proposed sources postdate the boot. | `/dev/sda` absent (disk flag not applicable); `sdb`–`sdf` disk flags not read; all 85 partition flags `ro=1` | `0x85fc0000–0x8a8fffff` reserved; `stmfts` exposed `event2` on I2C14; physical input included contact, changing X/Y coordinates and `ABS_MT_TRACKING_ID=-1` release. `qcom_wdt` was bound; feeder state: not read. Exact touch X/Y extrema were not retained. No UFS write. |
| 2026-11-12 (superseded) | `e37196952fca14d5a9152871927039b2ac1d311ec500710f8611f9ac6eff8072` | The booted test overlay itself defined the full touch node; its exact source was not retained. The later rebuild uses #249's disabled node plus [`device/dipper/overlays/fts5-touchscreen.dtso`](overlays/fts5-touchscreen.dtso), blob `f09381d59a110d02a8b8819f585e304efb1870a3`; both proposed sources postdate the boot. The exact source commit for the stale `base-panel.dtb` was not retained. | `/dev/sda` absent (disk flag not applicable); `sdb`–`sdf` disk flags not read; all 85 partition flags `ro=1` | Touch events worked, but the DTB omitted `removed-region@86200000` and `removed-region@89100000`; `/proc/iomem` exposed those areas as System RAM. This was noticed when reviewing `/proc/iomem` after the boot's touch test. Replaced by the corrected image above. No UFS write. |
| 2026-11-11 | `665fcbd682cd8e64b3ffec1ff7f4ab9ca33a26b9b47b11073be7a3c267abf2b1` | #505 `2d67708e07728b13b5d2a9eda299ace8a1608163`; kernel/DTS #220 `26aa77c317cd029d16e58890c532562ca275c9f0` | Node absent, disk `ro=0`; all 85 partitions `ro=1`; `sdb`–`sdf` `ro=1` | `uname -r` `7.3.0-rc6-omamobile-push1`; `qcom_wdt` bound and feeder running; simpledrm fb0, USB/SSH and reserved `0x85fc0000–0x8a8fffff` passed. Physical screen, keys, touch, GPU and Wi-Fi not tested. RAM boot only; no UFS write. |
| 2026-11-10 | `540b5d31ea73a31ade9a8d8cb5152cf701dc3c2ad8256734790b0ef2db6dbdd0` | Combined #503/#505; kernel/DTS #220 `055a4019998f22cc0e2a8afef7650274a6ce0657` | Node absent, disk `ro=0`; all 85 partitions `ro=1`; `sdb`–`sdf` `ro=1` | Watchdog/feeder, simpledrm, USB/SSH, reserved TZ range, pinned host key and `reboot-to bootloader` passed; physical keys not captured. |
| 2026-11-10 | `a2d435a7430e7e4c145c61cdb9866d16929731859d91c7105e6bfcaf9e3e9dd7` | Rebased #503 NFS image with #220 `055a4019998f22cc0e2a8afef7650274a6ce0657` | `/dev/sda` present, set `ro=1`; six disks and all 85 partitions read back `ro=1` | `uname`, NFS root, multi-user systemd, watchdog and SSH passed; generic Arch test root only; link-loss recovery and Omarchy desktop untested. |
Expand All @@ -27,6 +30,43 @@ name list. The retained backup and kernel records cover 84 non-userdata GPT
partitions plus `userdata`; the other seven names cannot be identified from the
retained artifacts and remain unknown.

## 2026-11-12 — corrected FTS5 RAM image

The first active-low FTS5 image (`e37196952fca14d5a9152871927039b2ac1d311ec500710f8611f9ac6eff8072`)
bound `st,stmfts5` on I2C14 and physical taps produced touch, multitouch X/Y
and release events inside the 1080×2248 panel. Its `base-panel.dtb` was stale:
the merged DTB omitted `removed-region@86200000` and
`removed-region@89100000`, exposing both ranges as System RAM in `/proc/iomem`.
The base DTB was reused from an earlier local build output; its exact source
commit was not retained. The missing reservation was noticed when reviewing
`/proc/iomem` after the boot's touch test, not before boot. The image is
superseded even though all 85 UFS partition flags were read-only.

Rebuilt the base DTB from the current preprocessed Dipper DTS, then reapplied
the same FTS5 overlay. The corrected RAM image
`1bf0ffd5302fff7fa75f49382567063d7e03bd52083baa81a88107a47dcbdbd2` booted on
kernel `7.3.0-rc6-omamobile-push1`. `/proc/iomem` showed
`0x85fc0000–0x8a8fffff` reserved; all 85 UFS partition flags were `ro=1`,
`/dev/sda` was absent, and `stmfts` plus `qcom_wdt` bound. The kernel reported
no Oops, BUG, panic or WARN. Physical touch on `/dev/input/event2` produced
`ABS_MT_TRACKING_ID`, changing `ABS_MT_POSITION_X/Y` values, and a release
event (`ABS_MT_TRACKING_ID=-1`). No UFS partition was written. The node is now
proposed disabled in kernel PR #249 at
`0d548847248512b1e8e1d40d66a85b81fe5ed471`; enabling it with the repository
overlay [`device/dipper/overlays/fts5-touchscreen.dtso`](overlays/fts5-touchscreen.dtso),
blob `f09381d59a110d02a8b8819f585e304efb1870a3`, rebuilds the tested node.
Both proposed sources postdate the historical boots.

Then booted a RAM image with the same FTS5 node enabled directly in the kernel
DTS. Image SHA-256:
`ac93a5d9f6787153cf620704f192c22da91cf2041789f16f4214111b8458d226`.
`stmfts` registered on `event2`; all 85 UFS partitions remained read-only,
`/dev/sda` was absent, and `/proc/iomem` retained the complete
`0x85fc0000–0x8a8fffff` reserved range. No Oops, BUG, panic or WARN appeared.
A physical touch event was not captured during this direct-DTS boot; the
preceding overlay boot with identical node properties captured contact,
changing X/Y and release. No UFS partition was written.

## 2026-11-11 — rebuilt #505 RAM image

Built and booted `boot.img` SHA-256
Expand Down Expand Up @@ -158,8 +198,9 @@ the owner confirmed the console was visible on the lit panel. All 85 partition
flags were `ro=1`; `/dev/sda` was absent, while `/dev/sdb` through `/dev/sdf`
remained present and had `ro=1` on this pre-#484 image. Backlight value 0 was
accepted and restored to 511 (maximum 1023). Framebuffer blanking changed
DRM `dpms` from `On` to `Off` and back to `On`. The FTS5 touch input still needs
a test on the current kernel. The panel log included a missing DSI `vdda`
DRM `dpms` from `On` to `Off` and back to `On`. FTS5 contact, coordinate
changes and release were later captured on 2026-11-12; see above. Hyprland
delivery remains untested. The panel log included a missing DSI `vdda`
supply (dummy regulator) and missing `a630_sqe.fw`; the display initialized,
but GPU rendering was not tested. The RAM image had no Arch root, and the probe
did not write UFS. The installed-system backup checksums were verified before
Expand Down Expand Up @@ -203,9 +244,9 @@ procedure and supported single-slot writer are still required before a flash.
fts,reset-gpio = <&tlmm 32 ...>; fts,default-fw-name = "st_fts_e1.ftb"; }`, with
`fts,tp-vendor = <0x48>`. The running kernel's input device is `N: Name="fts"`.
An `atmel,mxt-ts@4a` node exists as the alternate vendor and is not the active one.
**Open:** mainline's ST driver is `st,stmfts` (`drivers/input/touchscreen/stmfts.c`),
a different driver from Xiaomi's downstream `st_fts`; the current FTS5 driver
has not been tested with Dipper's DTS node. No sibling board helps —
Mainline's ST driver is `st,stmfts` (`drivers/input/touchscreen/stmfts.c`),
a different driver from Xiaomi's downstream `st_fts`; tested on Dipper
2026-11-12, see above. No sibling board helps —
polaris uses `syna,rmi4-i2c@20`, beryllium a FocalTech/Novatek part.
- **Audio** `sdm845-tavil-snd-card` (WCD9340). **Fingerprint** Goodix, capacitive
(`fingerprint_goodix`, `uinput-goodix`); no mainline driver, out of scope.
Expand Down Expand Up @@ -289,7 +330,9 @@ light was built in (`=y`).
### Touch: first failed probe (historical)

The original FTS4-only driver bound and timed out. These observations describe that
kernel revision; they are not results from the current FTS5 driver:
kernel revision; they are not results from the current FTS5 driver. The current
driver and Dipper node were tested on 2026-11-12; see the corrected FTS5 RAM image
above:

- `i2c-14` and `i2c-5` come up; the touch device **exists as `14-0049`**, so the bus and
address in the DTS are right.
Expand All @@ -314,7 +357,8 @@ At the time, it was unknown whether GENI/GPI or the FTS5 protocol caused the tim
The live controller is FTS5. A separate historical RAM boot on 2026-10-30, using
the unpublished kernel release above, registered `stmfts` and a physical tap
produced `ABS_MT_POSITION_X`. That observation is not a test of the current
kernel PRs. The public successor series and its binding are in
kernel PRs. The current driver and Dipper node were tested on 2026-11-12; see
the corrected FTS5 RAM image above. The public successor series and its binding are in
[kernel PR #219](https://github.com/derekross/linux-omarchy-mobile/pull/219).
The upstream v8 cover letter states it was tested on Xiaomi Mi 8:
[FTS5 v8 cover letter](https://lists.openwall.net/linux-kernel/2026/09/25/2211),
Expand Down
7 changes: 7 additions & 0 deletions device/dipper/overlays/fts5-touchscreen.dtso
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
// SPDX-License-Identifier: MIT
/dts-v1/;
/plugin/;

&touchscreen {
status = "okay";
};