Skip to content

arm64: dts: exynos: google: zuma-husky: Bluetooth on the APM UART - #199

Merged
derekross merged 7 commits into
derekross:omamobilefrom
Somnius:omamobile-zuma-husky-bt
Nov 7, 2026
Merged

derekross merged 7 commits into
derekross:omamobilefrom
Somnius:omamobile-zuma-husky-bt

Conversation

@Somnius

@Somnius Somnius commented Nov 4, 2026

Copy link
Copy Markdown
Contributor

Bluetooth on the Pixel 8 Pro. It comes up, pairs and plays audio.

The chip is the BCM4398D0, the Bluetooth half of the Wi-Fi combo, on the APM's USI1 UART (0x155d0000, IRQ 71). Everything comes from Google's GPL zuma tree (zuma-usi.dtsi, zuma-board-bluetooth.dtsi, zuma-pinctrl.dtsi, cmucal-sfr.c), built the way the 9a, the 7 Pro and the 6 Pro already do it.

4 commits:

  1. dt-bindings: clock IDs for the CMU_APM USI1 UART and SYSREG, plus google,zuma-apm-sysreg and google,zuma-uart.
  2. clk-zuma: the divider (0x1818) and the gates (0x20f4, 0x20f8, 0x2148, 0x2168), read-only like the rest of CMU_APM. The offsets are the same as Google's zuma cal-if and as clk-zumapro.c.
  3. zuma.dtsi: sysreg_apm, usi_apm1, uart18, all disabled.
  4. zuma-husky.dts and zuma-pinctrl.dtsi: BT_REG_ON gpn8-0, BT_DEV_WAKE gpn9-0, BT_HOST_WAKE gpn6-0, the UART pads gpa3-0 to gpa3-3 (function 2) and the REG_ON latch clock gpp24-3 (brcm,ble-dbo-gpios, which Bluetooth: hci_bcm: Add the BCM4383 and its REG_ON latch (Pixel 9a) #94 already handles for the 9a).

Two things were needed that I missed at first: without the pads and without the latch the chip never answered (command 0xfc18 tx timeout, then Reset failed -110). The compatible is brcm,bcm4389-bt, since hci_bcm has no BCM4398 entry; it only adds the 30 ms wait.

Tested on my 8 Pro with 7.3.0-rc6 (the e1 kernel, so Brett's tag plus my own commits, not plain omamobile), with the stock BCM.hcd installed as brcm/BCM.google,zuma-husky.hcd from my own phone's vendor image (not in the repo). dmesg: BCM4398D0 G5SN_V17_HK3 REL BR:0169 FW:b8718197dc, hci0 up, discovery finds devices, and my Soundcore A40 headphones paired and played YouTube in Chromium over A2DP.

On the current omamobile branch the series applies and clk-zuma.o and zuma-husky.dtb build, but I did not boot that exact tree, and I did not run dt_binding_check / dtbs_check.

Not tested: calls (HFP), mice and keyboards, suspend, power draw. The controller keeps a placeholder address until userspace sets the one in /chosen/config/bt_addr (I tested that by hand with the 7 Pro's omamobile-bt-addr; the husky device package needs it, separate PR). Whether the BCM4389 entry is the right one for the BCM4398 long term is open.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @Somnius, this is a lovely, careful PR. I checked every value against Google's tree. The UART at
0x155d0000 on interrupt 71, the sysreg at 0x154204e0 with offset 4, the CMU_APM divider and gates (offsets, bit
widths and parents, from cmucal-sfr.c and cmucal-node.c), and the Bluetooth pins with their pulls and drive
all match. The pin groups sit in the right controllers. Our gate passes on today's omamobile, merged with #200
as well, and the 8 Pro's DTB validates with no new messages. Each commit builds on its own too.

A few things before it goes in:

  1. Please name the chip in the compatible. A compatible is an ABI promise about the part, and this one is a
    BCM4398. Since hci_bcm's data for it would be the BCM4389's, the usual form is a fallback with no driver
    change: compatible = "brcm,bcm4398-bt", "brcm,bcm4389-bt";, and in brcm,bluetooth.yaml an items: entry
    with brcm,bcm4398-bt followed by const: brcm,bcm4389-bt, as it already has for infineon,cyw43439-bt.
    That's one small dt-bindings: net: bluetooth: brcm: commit before the board commit.
  2. The test and the tree that lands. Could you add the uname -r of the e1 kernel, the blob ids of
    zuma-husky.dts, zuma.dtsi, zuma-pinctrl.dtsi and clk-zuma.c in that tree
    (git rev-parse HEAD:<path>), and where its tag can be fetched? A tools/phone-smoke report from that boot
    would help too. If you can boot this branch on omamobile once, even better, since it carries everything
    Bluetooth needs.
  3. arm64: dts: exynos: google: zuma-husky: Bluetooth on the APM UART #199 and arm64: dts: exynos: google: zuma-husky: haptics and the rear torch #200 touch the same lines. They conflict in samsung,exynos-sysreg.yaml (keep both new
    compatibles, google,zuma-apm-sysreg then google,zuma-peric0-sysreg) and at the end of zuma-husky.dts
    (keep both blocks). I'll merge this one first, and arm64: dts: exynos: google: zuma-husky: haptics and the rear torch #200 then needs a rebase. The combined tree passes our checks,
    so it's mechanical.

Smaller, optional ones: upstream will want the bindings commit split by subsystem (dt-bindings: serial: samsung: … and dt-bindings: soc: samsung: exynos-sysreg: … next to the clock header). The clk commit still
says "Not run on a phone yet", which could now say it was tested with the board commit on top. And checkpatch
asks for the uart18_bus comment line in zuma-pinctrl.dtsi to be wrapped.

Thanks again. Bluetooth audio on the 8 Pro is a big step!

…SYSREG clocks

The Pixel 8 Pro's Bluetooth chip sits on the APM's USI1 UART at
0x155d0000. Its divider and gates are at the same CMU_APM offsets as
Google's GPL zuma cal-if (cmucal-sfr.c: DIV_CLK_APM_USI1_UART 0x1818,
IPCLK 0x20f4, PCLK 0x20f8, RSTNSYNC 0x2148, SYSREG_APM 0x2168).

Signed-off-by: Lef <me@lefteros.com>
The sysreg of the Tensor G3's APM block, which holds the USI mode
selectors; the Pixel 8 Pro's Bluetooth UART needs it.

Signed-off-by: Lef <me@lefteros.com>
The zuma UART, for the APM USI1 UART that the Pixel 8 Pro's Bluetooth
chip sits on (0x155d0000).

Signed-off-by: Lef <me@lefteros.com>
Read-only, as the rest of CMU_APM. Offsets from Google's GPL zuma cal-if
(cmucal-sfr.c), which match clk-zumapro.c's for the same blocks: USI1
UART divider 0x1818 (7 bits, parent the APM function gate), IPCLK 0x20f4
and RSTNSYNC 0x2148 behind it, PCLK 0x20f8 and SYSREG 0x2168 on the NOC
divider.

Tested on a Pixel 8 Pro with the Bluetooth board commit of this series
on top: the UART comes up and the controller answers.

Signed-off-by: Lef <me@lefteros.com>
usi_apm1 at 0x155d00c0 with uart18 at 0x155d0000, interrupt 71 (Google's
zuma-usi.dtsi, IRQ_APM_USI1_UART_INT_ALIVE), on the pattern of zumapro's
and of USI9 here. Both stay disabled; a board enables them.

Signed-off-by: Lef <me@lefteros.com>
The Pixel 8 Pro's Bluetooth is a BCM4398. hci_bcm has no data of its
own for it and uses the BCM4389's, so the compatible names the chip and
falls back to brcm,bcm4389-bt, the same form as infineon,cyw43439-bt.
No driver change is needed.

Also mention the BCM4398 next to the Pixel 9a's BCM4383 in the
description of brcm,ble-dbo-gpios, as it latches BT_REG_ON the same way.

Signed-off-by: Lef <me@lefteros.com>
The Pixel 8 Pro's Bluetooth is the BCM4398D0 on the APM's USI1 UART
(uart18, 0x155d0000), as in Google's GPL zuma-board-bluetooth.dtsi:

 - BT_REG_ON gpn8-0, BT_DEV_WAKE gpn9-0, BT_HOST_WAKE gpn6-0 (rising
   edge);
 - the UART pads gpa3-0 to gpa3-3 in function 2 (Google's uart18_bus);
 - gpp24-3 as brcm,ble-dbo-gpios: BT_REG_ON reaches the chip through a
   flip-flop clocked by it, as on the 9a, and hci_bcm already clocks
   it.

Without the pads or the latch the chip never answers the first command
(0xfc18 tx timeout, then Reset failed -110). The compatible names the
BCM4398 with the BCM4389 as fallback, as hci_bcm has no BCM4398 entry;
the fallback only adds the 30 ms wait.

Tested on a Pixel 8 Pro with 7.3.0-rc6 and the stock BCM.hcd installed
as brcm/BCM.google,zuma-husky.hcd: hci0 comes up as "BCM4398D0
G5SN_V17_HK3 REL BR:0169 FW:b8718197dc", discovery finds devices, and a
pair of Soundcore A40 headphones paired and played YouTube over A2DP.
Not tested: calls (HFP), mice and keyboards, suspend, power draw. The
controller keeps a placeholder address until userspace sets the one in
/chosen/config/bt_addr.

Signed-off-by: Lef <me@lefteros.com>
@Somnius

Somnius commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

@derekross thank you. Reworked on today's omamobile (29a7a93), it's on my fork as zuma-husky-bt-v2 for now (https://github.com/Somnius/linux-omarchy-mobile/tree/zuma-husky-bt-v2), and I'll move this PR's branch onto it next. That rewrites the PR branch's history, so I'll say so here when I do.

What changed: the bindings are split by subsystem (the clock header, samsung,exynos-sysreg, samsung_uart), there's a dt-bindings: net: bluetooth: brcm: commit before the board one with the items: entry brcm,bcm4398-bt then const: brcm,bcm4389-bt, and the board file uses "brcm,bcm4398-bt", "brcm,bcm4389-bt". The clk commit now says it was tested with the board commit on top, and the uart18_bus comment is wrapped, plus another comment line in zuma-pinctrl.dtsi that checkpatch counted over 100 columns.

tools/kernel-pr-check passes on the 7 commits (merges cleanly, signed off, checkpatch --strict clean, W=1 build, 10 device trees build). dt-validate on the merged husky dtb prints only the two eusb2 phy supply messages that are already on omamobile (I could not run the full dt_binding_check here: the yamllint package is missing, only dt-doc-validate on the three yaml files, which is clean).

Still to add, from a boot of the combined #199 and #200 tree on omamobile: the uname -r, the four blob ids, the tag and a phone-smoke report. I built that tree (husky-pr-test on my fork, both conflicts kept both sides) but haven't booted it yet, it needs me at the phone. What I tested so far is on my e1 tree, which is public now: tag husky-test-20261106-e1-aoc on my fork.

@Somnius
Somnius force-pushed the omamobile-zuma-husky-bt branch from 8664827 to 69977ef Compare November 6, 2026 05:07
@Somnius

Somnius commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

I force-pushed this branch (now 69977ef) with the reworked series from my last comment: split bindings, the brcm bcm4398-bt binding before the board commit, wrapped comments. Same content otherwise. Boot test of the combined tree with #200 is still to come.

@Somnius

Somnius commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

I booted the combined tree (origin/omamobile + #199 + #200, merge b0fb5bc, tag husky-test-20261106-pr199-200 on my fork) on my 8 Pro. uname -r 7.3.0-rc6-omamobile-push1-gb0fb5bccb3d6, it was up 223 s when I ran tools/phone-smoke and all the required checks passed.

Blob ids at that commit:

  • arch/arm64/boot/dts/exynos/google/zuma-husky.dts b3fcd8716155f6fdc877cffe4e386ed2b33fb6b8
  • arch/arm64/boot/dts/exynos/google/zuma.dtsi 21c184ae870ca6cf83466456dac5a28089b76228
  • arch/arm64/boot/dts/exynos/google/zuma-pinctrl.dtsi 74b69056313b2dbb602e513df5fe86f1e7227106
  • drivers/clk/samsung/clk-zuma.c 833d676e8b92cf5e600604a7db87d65ed39b94dc

Bluetooth side: hci0 shows up and phone-smoke says powered true. I only checked that it registers and powers on. dmesg also has a lot of hci0: Failed to read codec capabilities (-22) lines, I haven't looked into those yet. I did not pair anything or test audio in this boot.

@derekross

Copy link
Copy Markdown
Owner

Thanks for the quick rework, @Somnius. I went through it point by point. The compatible now names the BCM4398 with the BCM4389 as its fallback, the bindings are split by subsystem, and the comments are wrapped. I fetched husky-test-20261106-pr199-200. It is today's omamobile with exactly this PR and #200 merged, and the four blob ids you posted match it. Our gate passes on today's base, CI is green including dtbs_check and dt_binding_check, and I validated the 8 Pro's DTB with this whole batch of your PRs merged together: no new messages.

One small thing before I merge: could you paste the tools/phone-smoke report from that boot into this thread? It names no phone, and it's the record we keep for every kernel PR.

Optional, if you touch it again: the edited brcm,ble-dbo-gpios description line in brcm,bluetooth.yaml is 108 columns (upstream wraps at 80). The board commit's "Tested" paragraph describes the A2DP test on the e1 tree, and adding a line that this series, with #200, was booted and hci0 registered and powered would make that clear. The Failed to read codec capabilities (-22) lines are harmless, by the way: the 7 Pro's BCM4389 prints the same ones.

This one goes in first, then #200 after its rebase. Thank you!

@Somnius

Somnius commented Nov 7, 2026

Copy link
Copy Markdown
Contributor Author

@derekross here is the tools/phone-smoke report from the combined tree boot (husky-test-20261106-pr199-200, kernel 7.3.0-rc6-omamobile-push1-gb0fb5bccb3d6, up 223 s, 2026-11-06T05:12Z). I didn't save it to a file that day, so I recovered it from my session log and I'm pasting it as it was, not re-running it. Two rows say "not ok": the GPU (no render node, this tree has no Panthor node) and the systemd units (omamobile-sound and omamobile-thermal, which are my own local units; I haven't looked at why they failed on that tree). Wi-Fi, sound and sensors are skipped because this tree has none of them. On Bluetooth I only checked that hci0 registers and powers on, no pairing and no audio on that boot.

Kernel 7.3.0-rc6-omamobile-push1-gb0fb5bccb3d6, tools at 722e530, 2026-11-06T05:12Z.

Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-push1-gb0fb5bccb3d6, slot b, up 223s
storage guard (required) ok 62 partitions; only userdata and slot B's boot images writable
root (required) ok ext4 on userdata, 206G free
kernel health (required) ok no oops or BUG; 0 WARN, 18 error lines in dmesg
systemd units not ok failed: omamobile-sound.service omamobile-thermal.service
USB gadget (required) ok bound to 11210000.usb
display ok card0-Unknown-1 connected, 1344x2992
GPU not ok no render node
session (required) ok Hyprland and the Omarchy shell running
lock ok PAM file present, the shell can lock (passwordPam true, locked true)
touch ok SALIXP Touchscreen
buttons ok gpio-keys
Wi-Fi skip no wireless interface
Bluetooth ok hci0, powered true
battery ok 92% Charging, 33 °C
sound cards skip none
sensors skip no sensor proxy
cameras skip no libcamera tools
thermal ok hottest zone 33 °C

@derekross

Copy link
Copy Markdown
Owner

Thank you, @Somnius. That's the report I wanted, and thanks for saying it was recovered rather than re-run. The release matches the test integration of this PR and #200 that I checked before, the boot comes after both heads, and the two "not ok" rows are the missing GPU node on that tree and your own local units, nothing from this series. I also checked that it still merges cleanly onto today's omamobile (with #205 in). Merging now; #200 can rebase onto it.

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