Skip to content

arm64: dts: exynos: google: zumapro-tegu: Add Bluetooth - #122

Merged
derekross merged 1 commit into
derekross:omamobilefrom
W4HO-ham:tegu-dt-bluetooth
Nov 1, 2026
Merged

derekross merged 1 commit into
derekross:omamobilefrom
W4HO-ham:tegu-dt-bluetooth

Conversation

@W4HO-ham

@W4HO-ham W4HO-ham commented Oct 28, 2026 •

Copy link
Copy Markdown
Contributor

Stacked on #106 (the DT stack); only the last commit is new. It booted in this form on the palmtop before the review changes (round bt2, below); the host-wake pull-down is new since, and a boot of this head follows.

The 9a's BCM4383 (driver: kernel #94, merged) sits on the APM's UART (USI1, uart18) at 3 Mbaud:

  • Power: BT_REG_ON on gpn8-0, behind the chip's latch clocked by gpp24-3 (brcm,ble-dbo-gpios, from Bluetooth: hci_bcm: Add the BCM4383 and its REG_ON latch (Pixel 9a) #94's binding).
  • Wake: device wake on gpn9-0, host wake on gpa5-6. hci_bcm takes host wake as its interrupt, which turns on its runtime PM and lets the controller wake the system.
  • gpa5: becomes an interrupt controller with its eight EINT SPIs (60-67).

The pins and values follow Google's GPL tegu Bluetooth DT and match what runs on the palmtop: hci0 up, BlueZ, A2DP, host wake.

Checks: checkpatch --strict clean; zumapro-tegu.dtb builds with W=1 on today's omamobile with the stack (rebased 2026-10-28 evening). The bt2 run used the integration tree's drivers/bluetooth/hci_bcm.c at blob 2d46b1da1991e7062a6e4f6555f0d9a7be11b26e, unchanged from that tree's HEAD.

@W4HO-ham

W4HO-ham commented Oct 28, 2026 •

Copy link
Copy Markdown
Contributor Author

Booted in this form on the palmtop (round bt2: the integration kernel 7.3.0-rc6-omamobile-tegu-s2mpg15-m1-g991042e2eff3-dirty, with usi_apm1, uart18 and the Bluetooth node enabled in the DT as this PR has them, not at runtime):

  • usi@155d00c0: okay. hci0 comes up on hci_bcm (btmgmt info: version 13, manufacturer 15 Broadcom), and bluetooth.service is active.
  • Host wake: gpa5 6 Edge host_wake counted 3 interrupts by the check, so the gpa5 EINTs and hci_bcm's runtime PM are wired.
  • USB gadget: still configured; nothing else on the board changed.
  • The log has "Failed to read codec capabilities (-22)": the controller doesn't support that optional HCI command, and hci_bcm carries on.

@derekross

Copy link
Copy Markdown
Owner

Thank you, this is a tidy node. I checked it against Google's GPL DT, which for Bluetooth is the zumapro repository's zumapro-board-bluetooth_wakectrl.dtsi, included from tegu's zumapro-tegu-common.dtsi. BT_REG_ON on gpn8-0, DEV_WAKE on gpn9-0, the latch clock on gpp24-3 (input, no pull, as stock), host wake on gpa5-6 and serial_18 all match. gpa5's SPIs 60-67 are stock's IRQ_ALIVE_EINT50-57. The brcm,ble-dbo-gpios dependencies are met, and the DTB builds without warnings and validates with no messages.

Three small changes, please:

  1. Comments split from their nodes. The &uart18 block went in between the storage-guard comment ("Behind the init's storage guard …") and &ufs, so that comment now sits above the Bluetooth comment. Please move the block above it. Separately, a blank line was added between the touchscreen comment ("Hold the touchscreen in reset …") and &usi2_hsi0. Please drop that one.
  2. Host wake needs its pull-down. Stock gives gpa5-6 EXYNOS_PIN_FUNC_EINT with EXYNOS_PIN_PULL_DOWN (bt_host_wake), and so does the Pixel 7 Pro's Bluetooth node in our tree. Here the pin has no pin group, so its pull is whatever the bootloader left. hci_bcm marks the device as a wakeup source, and the interrupt is rising-edge, so a floating line could wake the phone from s2idle. Please add a bt-host-wake-pins group, with EINT and pull-down, to pinctrl-0 (gpa5 is an alive bank, so no -pdn properties).
  3. The message's citation. "Google's GPL tegu Bluetooth device tree" is actually that zumapro file. A Link: to it at the commit you used would make it checkable.

Two notes on the description: it says "Draft", but the PR isn't one, and it says the DTB builds "on today's omamobile". It doesn't merge there any more (the Makefile conflict on #80), so CI hasn't run. The bt2 run was on an integration tree that isn't published (g991042e2eff3-dirty). That's fine for a board node whose driver is already merged, but please add the blob id of the drivers/bluetooth/hci_bcm.c you ran (git rev-parse HEAD:drivers/bluetooth/hci_bcm.c), so it can be tied to the driver that's on omamobile.

This lands after #106 (it's stacked on it) and after the rebase on #80.

@W4HO-ham

Copy link
Copy Markdown
Contributor Author

All three are in 20f58cf, rebased on #106's new head:

  1. The &uart18 block now sits above the storage-guard comment, so that comment is back over &ufs, and the stray blank line after the touchscreen comment is gone.
  2. Host wake has its pin group: bt_host_wake (gpa5-6, EINT, pull-down, no -pdn, as stock's bt_host_wake) in &pinctrl_far_alive, added to the node's pinctrl-0.
  3. The message cites Google's zumapro zumapro-board-bluetooth_wakectrl.dtsi with a Link: at 8831630eee2a, the revision I compared against.

On the description: it's not a draft, and since this morning's rebase it does sit on today's omamobile. The bt2 run used the integration tree's drivers/bluetooth/hci_bcm.c at blob 2d46b1da1991e7062a6e4f6555f0d9a7be11b26e, which is unchanged from its HEAD (no local edits). I'll update the description to say so. The pull-down is new since that boot, so I'll boot this head once the palmtop is back and post the result.

@derekross

Copy link
Copy Markdown
Owner

Thanks, all three are in, and I checked each one against 20f58cf4b74c:

  1. The &uart18 block now comes after &sysreg_ufs, so the storage-guard comment sits directly over &ufs again. The stray blank line above &usi2_hsi0 is gone.
  2. bt_host_wake (gpa5-6, EINT, pull-down, no power-down properties) is in &pinctrl_far_alive, which is where zumapro.dtsi puts the gpa5 bank, and it's in the node's pinctrl-0. It matches stock's bt_host_wake in zumapro-board-bluetooth_wakectrl.dtsi at the revision you linked. Leaving out the -pdn properties is right for an alive bank.
  3. The Link: resolves and points at that file, the trailers parse, and the message now names the zumapro DT and the pull-down.

Apart from the rebase onto #106's new head, this is the only commit in the stack that changed. The stack merges cleanly onto today's omamobile (after #113 and #123). CI passes on this head, and my local gate passes on the combined top (#118, #119 and this PR together): sign-offs, prefixes, checkpatch, W=1 builds and the device trees. The other Google boards' DTBs are byte-identical to the base.

Before I merge, I need the boot of this head that you mentioned. One detail matters for it. Blob 2d46b1da1991 (the hci_bcm.c from the bt2 run) isn't in this tree. The hci_bcm.c that lands with this PR is 2b1b47550e7b (today's omamobile, with #94), so bt2 ran a different driver as well as the old pins. Please boot a kernel built from this head, or from a tree whose git rev-parse HEAD:drivers/bluetooth/hci_bcm.c gives 2b1b47550e7b, and post:

  • uname -r and the commit (or that blob id);
  • the tools/phone-smoke report (its Bluetooth row);
  • hciconfig -a or bluetoothctl show with the address left out, plus the dmesg lines for hci_bcm/Bluetooth: hci0;
  • the host-wake line from /proc/interrupts (the gpa5 EINT), read once with Bluetooth idle and again after some traffic (a pairing scan or a short A2DP play), so we can see the pin isn't firing on its own with the pull-down;
  • if it's easy in the same boot, dmesg | grep -i ramoops and ls /sys/fs/pstore (for arm64: dts: exynos: google: zumapro: Enable ramoops #106).

You don't need a suspend test for this. Since the pull-down is the only change in the pins, the bt2 results plus this boot are enough.

On order: this lands after #106 (and so after #96 and #105). It doesn't conflict with #119 or #118.

Small thing: the PR description still says "Google's GPL tegu Bluetooth DT", while the commit now says zumapro. Only the commit lands, so it's cosmetic.

@W4HO-ham

Copy link
Copy Markdown
Contributor Author

Rebased onto #106's new head (one commit, unchanged). Compiles; checkpatch clean.

@derekross

Copy link
Copy Markdown
Owner

Thanks for the rebase. The node is unchanged and still fine. What I still need is the boot of this head from 10-29: the Bluetooth row, the gpa5 host-wake count idle and after some traffic, and the hci_bcm.c blob. It can sit on #96 rather than #105 (it picks cleanly there now that ramoops is in).

@W4HO-ham

Copy link
Copy Markdown
Contributor Author

Re-stacked onto #96 as suggested; it picked cleanly with ramoops in omamobile. One commit of its own, the node unchanged. Device tree compiles, checkpatch clean. The boot of this head for the Bluetooth row and the host-wake counts is still owed.

@W4HO-ham

Copy link
Copy Markdown
Contributor Author

The boot of this head (6ec07ae), with the host-wake pull-down, on the palmtop. It's built the same way as #96's run today: defconfig, omamobile.config with omarchy-mobile#227's four USB/PMIC options, and olddefconfig, with the same bootargs overlay. uname -r is 7.3.0-rc6-omamobile-push1-g6ec07aeee91d, no -dirty. The DTB sha256 before the overlay is fe47bfc1da338beaf6db24170e4abc51ae980e4c45f58f922a16a4adf3f550a9.

  • usi@155d00c0: okay. bluetooth.service active, and hci0 up on hci_bcm (btmgmt info: version 13, manufacturer 15, Broadcom). tools/phone-smoke: | Bluetooth | ok | hci0, powered true |.
  • Host wake, from /proc/interrupts (gpa5 6 Edge host_wake): 6 when idle about 20 s after bluetoothd came up, and 38 after a 12 s bluetoothctl scan on. So it counts with traffic and stays quiet without it, now with the pull-down.
  • drivers/bluetooth/hci_bcm.c in this tree: blob 2b1b47550e7b2b7ae6fd8cb79fd98eb2e9126ba6 (git rev-parse HEAD:drivers/bluetooth/hci_bcm.c at 6ec07ae). The earlier bt2 run's integration-tree blob was 2d46b1da1991…; this one is what's on omamobile under the stack.
  • The only Bluetooth lines in dmesg --level=err are "Failed to read codec capabilities (-22)", repeated as before: the controller doesn't implement that optional command, and hci_bcm carries on.

The palmtop is back on its daily kernel.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @W4HO-ham. That's everything I asked for on the 29th, from a kernel built from this exact head:

  • 7.3.0-rc6-omamobile-push1-g6ec07aeee91d has no -dirty, and its hash is this PR's head.
  • drivers/bluetooth/hci_bcm.c at this head is 2b1b47550e7b, the same blob as today's omamobile. So the boot ran the driver that lands with this PR, not the integration tree's.
  • The smoke test's Bluetooth row is ok (hci0 powered) and btmgmt reports the Broadcom controller.
  • The host wake reads 6 when idle and 38 after a 12-second scan, so with the pull-down the pin counts with traffic and stays quiet without it. That was the point of the check.
  • "Failed to read codec capabilities (-22)" is the controller not implementing an optional command. hci_bcm carries on, as you say.

The node is unchanged from the version I checked. The gate passes merged onto today's omamobile (sign-offs, checkpatch with no warnings, eight device trees with no new warnings), the 9a's DTB passes dt-validate with no messages, and CI is green. It also merges cleanly with #119, #136 and #105: I built the 9a's DTB with all four combined, and it validates with no messages.

So it's ready. It lands after #96. Once #96 is in, a git rebase omamobile leaves just this commit, and I'll merge it then.

The Pixel 9a's BCM4383 sits on the APM's UART (USI1, uart18) at 3 Mbaud.
BT_REG_ON is gpn8-0 behind the chip's latch, clocked by gpp24-3 (the
binding's brcm,ble-dbo-gpios), device wake is gpn9-0 and host wake is
gpa5-6 (EINT with a pull-down, as stock's bt_host_wake), as in Google's
GPL zumapro Bluetooth device tree that tegu includes. With host wake as
its interrupt, hci_bcm runs the controller with runtime PM and lets it
wake the system.

gpa5 becomes an interrupt controller with its eight EINT SPIs (60-67).

Link: https://android.googlesource.com/kernel/devices/google/zumapro/+/8831630eee2ad095b85ed452944e6353940966a5/dts/zumapro-board-bluetooth_wakectrl.dtsi
Signed-off-by: W4HO <87742609+W4HO-ham@users.noreply.github.com>
@W4HO-ham

W4HO-ham commented Nov 1, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto today's omamobile (with #96); details on #119.

@derekross
derekross merged commit f43a44c into derekross:omamobile Nov 1, 2026
1 check passed
@derekross

Copy link
Copy Markdown
Owner

Thanks, @W4HO-ham. The rebase changed only the base: the commit is the same patch as 6ec07aeee91d, message included, so the boot of that head (no -dirty, hci_bcm.c the same blob as omamobile) still covers it. The gate passes merged onto today's omamobile, CI is green with dtbs_check, and it also applies cleanly on top of #161. Merged with "Rebase and merge".

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants