Skip to content

arm64: dts: exynos: google: gs201-lynx: enable Bluetooth on the APM UART - #98

Merged
derekross merged 1 commit into
derekross:omamobilefrom
FlorencioJP:lynx-bluetooth
Nov 3, 2026
Merged

derekross merged 1 commit into
derekross:omamobilefrom
FlorencioJP:lynx-bluetooth

Conversation

@FlorencioJP

Copy link
Copy Markdown

Bluetooth for the Pixel 7a (lynx): the WCN6855 BT core on the APM USI1 UART, on the same shape as cheetah's BCM4389 block.

  • cmu_apm, usi_apm1, serial_apm1, pinctrl_peric0 enabled; bluetooth child qcom,wcn6855-bt, BT_EN gpp16-2, the Wi-Fi/BT 0.95 V buck on gph2-5 as a fixed regulator with a 7 ms startup delay (stock btpower order). WLAN_EN stays low.
  • max-speed = <3000000>: 3.2 Mbaud fails every TLV segment (Frame reassembly failed (-84) then command 0xfc00 tx timeout); 1, 2 and 3 Mbaud work.
  • Tested on the phone with hci_uart/btqca and linux-firmware-atheros (qca/hpbtfw21.tlv, hpnv21.bin): SOC version 0x400c0210, firmware loaded at 3 s, Powered: yes, scan lists the neighbours. The controller has no public address of its own; userspace sets the bootloader's /chosen/config/bt_addr, as cheetah's omamobile-bt-addr does (userspace side in derekross/omarchy-mobile, separate PR).
  • Enabling cmu_apm by runtime overlay oopses (gs201_cmu_probe is __init), so the nodes live in the boot DTB.

Context and logs: derekross/omarchy-mobile#36 (comment 6455522265).

@derekross

Copy link
Copy Markdown
Owner

Thank you, @FlorencioJP. Bluetooth on a second gs201 phone is very welcome. The commit message already holds the two things I'd most want written down: the 3.2 Mbaud failure, with 1, 2 and 3 Mbaud working, and why the cmu_apm nodes have to be in the boot DTB (gs201_cmu_probe is __init). The pins match Google's gs201-lynx-bluetooth.dtsi, and the change stays in gs201-lynx.dts, so nothing changes for the other gs201 phones.

Some things need sorting out before it can merge:

  1. It doesn't build on omamobile. dtc stops at the new regulator:

    gs201-lynx.dts:123.34-129.4: ERROR (phandle_references): /regulator-bt-buck: Reference to non-existent node or label "gph2"
    

    gs201's HSI2 pin controller (0x14440000, the gph2 and gph4 banks) isn't on the published branch yet. It lives in my integration tree, in the commit that adds the Pixel 7 Pro's PCIe host. The base the description names (ccf09ecce05d) doesn't have it either, so the phone must have run a tree with something extra. Please add its uname -r and say where gph2 came from. I'll publish the pin controller on its own (the Pixel 7 Pro already runs it), and then this can rebase onto it. With that node added, I checked that the rest of your DTS compiles.

  2. The sign-off. The commit's author is Version Tracker <…@users.noreply.github.com>, and the Signed-off-by: is FlorencioJP <claude@version-tracker.app>. The kernel needs the same human identity as author and as signer (the DCO). Our check also reads a claude@ address as an AI agent's sign-off. If an assistant helped, credit it with Assisted-by:, as Documentation/process/coding-assistants.rst on omamobile describes, and sign off with your own name and address. git commit --amend --reset-author with your identity set, then the new trailer, does it.

  3. The binding's required supplies. qcom,wcn6855-bt.yaml requires vddaon, vddbtcmx, vddrfa0p8, vddrfa1p2, vddrfa1p7, vddrfacmn, vddwlcx and vddwlmx, and marks enable-gpios and vddbtcxmx-supply as deprecated. Once the file builds, dtbs_check will report the missing eight, and CI requires no new warnings. Upstream describes this chip with a qcom,wcn6855-pmu node (qcom,qca6390-pmu.yaml, driver pwrseq-qcom-wcn.c). It holds bt-enable-gpios and wlan-enable-gpios and takes the buck as an input supply, and the Bluetooth node takes its supplies from the PMU's outputs, with no enable-gpios. The PMU's xo-clk-gpios is a short pulse, so gph2-5 would stay a fixed regulator feeding the PMU, not that property. That's more work than this patch, but it also solves the next point. I'm happy to look at a draft before you spend a boot on it.

  4. Turning Bluetooth off cuts the Wi-Fi buck. On a WCN6855, hci_qca's power-off (qca_power_off()) drives BT_EN low, waits 100 ms and disables every supply. With the fixed regulator as the only consumer, that drops gph2-5. Google's btpower never drops that pin when Bluetooth turns off. It also happens at every boot: a controller with no address is unconfigured, and hci_power_on() closes it right after the firmware download, so the buck goes up, down, and up again once your service sets the address. Nothing breaks today, because Wi-Fi is parked. But once Wi-Fi runs, Bluetooth off would take it down. And this node will be in every lynx's kernel, latentoperator's Wi-Fi test phone included: a Wi-Fi test branch that drives gph2-5 itself will collide with the regulator. A PMU node shared by both radios avoids all of this. If you keep the fixed regulator for now, please say in the commit message that Bluetooth owns gph2-5 and drops it on power-off.

  5. Evidence. Please add the tools/phone-smoke report and uname -r from the phone, and name the plan or NOTES section the test follows (arm64: dts: exynos: google: gs201-lynx: enable Bluetooth on the APM UART #98 in omarchy-mobile has the facts). It was tested on one chip revision, 400c0210. latentoperator's lynx reports 400c1211, and by btqca's naming that revision loads a different NVM file (the g variant, hpnv21g.bin or a board-specific one). So one boot on that phone, or "untested on the 400c1211 revision" in the message, would cover it.

Smaller things: one line of the message is 76 characters (checkpatch flags it). The node has no device-wake or host-wake line; that's fine for hci_qca, but it means Bluetooth can't wake the phone, so please say so.

On order: #93 (touch) also appends to the end of gs201-lynx.dts, and it's ready, so it will land first. Rebasing onto it then gives a small conflict at the end of the file, and keeping both blocks resolves it. I checked that the combined file compiles. The CI run for this PR is waiting for my approval. I'll approve it, but expect it to show the build failure above until the pin controller is published.

@FlorencioJP

Copy link
Copy Markdown
Author

Thanks for the review. Done on the branch (same tree, new commit f58642f0):

  • Sign-off. Author and signer are now Jean-Paul Florencio jpflorencio@version-tracker.app, with Assisted-by: LLM as coding-assistants.rst describes.
  • Where gph2 came from. The phone ran 7.3.0-rc6-omamobile-lynx-wifi1, built from your integration tree at 1cb4bd36f3be (the tree with the HSI2 pin controller), not from ccf09ecce05d alone; the message now says so. I will rebase onto the pin controller once you publish it, and onto arm64: dts: exynos: google: gs201-lynx: enable the touchscreen #93.
  • Message. Bluetooth owns gph2-5 and drops it on power-off (and once per boot before the address is set), no wake lines so Bluetooth cannot wake the phone, tested on 0x400c0210 only and untested on 0x400c1211, all lines under 76 characters.
  • PMU node. Agreed that a shared qcom,wcn6855-pmu node is the right shape, and it fixes the Wi-Fi buck problem for latentoperator's phone too. I will send a draft for a look before spending a boot on it, after the pin controller lands.
  • Evidence. The tools/phone-smoke report is pending: the phone is off the bench right now; it will go in the omarchy-mobile#98 notes with the next boot.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @FlorencioJP, for the quick turnaround. The new commit fixes the identity: author and signer are both you, the Assisted-by: LLM line matches coding-assistants.rst on omamobile, and our check now passes the sign-off. checkpatch has nothing left to say, and the message is a good record. It names the kernel the phone ran (I checked that 1cb4bd36f3be is the published integration tree), says Bluetooth owns gph2-5 and drops it on power-off, says there are no wake lines, and says only 0x400c0210 was tested. I compared the trees: the device tree is byte-for-byte the one I reviewed, so the rest of my first review still applies.

Where things stand:

  1. The pin controller is on me. gph2 still isn't on omamobile, so this still stops at dtc. Until I publish the HSI2 pin controller there's nothing for you to do on this point. To check the rest in the meantime, I added that controller to today's omamobile by hand and resolved the conflict with arm64: dts: exynos: google: gs201-lynx: enable the touchscreen #93 (now merged). Keep &usi0 { … }; from omamobile and put your Bluetooth block after it. The board then builds with W=1 and no new warnings. When you rebase, that's the whole resolution.

  2. The PMU node needs to be in this PR, not in a follow-up. In my first review I wrote "if you keep the fixed regulator for now, please say so in the commit message". I shouldn't have offered that. Our contributing guide asks for dtbs_check with no new warnings, and as soon as this file builds, qcom,wcn6855-bt.yaml will report the eight required supplies the node doesn't have. So the fixed-regulator version can't pass CI on its own. I'm sorry for the detour. Your plan to send a qcom,wcn6855-pmu draft is the right next step, and I'm happy to look at it before you spend a boot on it. The pieces are already in our config (CONFIG_POWER_SEQUENCING_QCOM_WCN=m comes from defconfig).

  3. A correction to what the PMU node fixes. I said a shared PMU node "avoids all of this". That's only true once Wi-Fi uses it as well. pwrseq-qcom-wcn reference-counts its regulators unit across the Bluetooth and Wi-Fi targets. With Bluetooth as the only user (lynx has no PCIe Wi-Fi node yet), switching Bluetooth off still disables the PMU's inputs, and with them the gph2-5 buck. So please keep the sentence about Bluetooth owning gph2-5 in the message for the PMU version too, until a Wi-Fi node takes the wlan target. One more thing for the draft: if the PMU node holds wlan-enable-gpios (gph2-4), the kernel owns that pin from boot. The Wi-Fi bench then can't drive it by hand, and its runbooks should say so. Leaving wlan-enable-gpios out until there's a PCIe node is also fine by the binding, and either choice is reasonable if the message says which one you picked.

  4. Evidence on the version that lands. The PMU version changes the power sequence (the pwrseq unit order instead of the regulator's 7 ms delay), so the earlier boot covers the UART and firmware side but not the new sequence. Please run the next boot, with the tools/phone-smoke report, uname -r, dmesg | grep -i -e qca -e bluetooth and bluetoothctl show, on the PMU version rather than on this head. That saves you a boot.

On order: I publish the pin controller, you rebase onto it with the PMU node, and once CI is green including dtbs_check, this merges. omarchy-mobile #98 follows, so its links point at merged work.

Thanks again. This is careful work, and the 3.2 Mbaud finding and the __init trap are exactly the kind of thing I'm glad to see written down.

@FlorencioJP

Copy link
Copy Markdown
Author

Draft of the qcom,wcn6855-pmu version, for a look before I spend a boot on it. It compiles with dtc against the integration tree at 1cb4bd36f3be (the one with gph2); I could not run dtbs_check here (no dtschema in my build container), so the binding side is by reading qcom,qca6390-pmu.yaml and qcom,wcn6855-bt.yaml.

What changes against the reviewed head: the fixed regulator stays, as the PMU's input, and the Bluetooth node loses enable-gpios and vddbtcxmx-supply. The PMU node owns gpp16-2 and the pinctrl entry for it. pwrseq-qcom-wcn's bluetooth target then does: regulators on (the buck, with its 7 ms startup-delay-us), BT_EN high, 50 ms (pwup_delay_ms); off is BT_EN low, then the regulators. So the order and the delay before BT_EN are the same as the reviewed version, and Bluetooth still owns gph2-5 and drops it on power-off until a Wi-Fi node takes the wlan target. The message will say so.

/ {
	lynx_wcn_buck: regulator-wcn-buck {
		compatible = "regulator-fixed";
		regulator-name = "lynx-wcn-0p95-buck";
		regulator-min-microvolt = <950000>;
		regulator-max-microvolt = <950000>;
		gpio = <&gph2 5 GPIO_ACTIVE_HIGH>;
		enable-active-high;
		startup-delay-us = <7000>;
	};

	/*
	 * The PMIC rails stock keeps on for the chip and the host never
	 * switches; the PMU binding needs a handle for each input.
	 */
	lynx_wcn_pmic: regulator-wcn-pmic {
		compatible = "regulator-fixed";
		regulator-name = "lynx-wcn-pmic-rails";
		regulator-always-on;
	};

	wcn6855-pmu {
		compatible = "qcom,wcn6855-pmu";
		pinctrl-0 = <&bt_reg_on>;
		pinctrl-names = "default";

		bt-enable-gpios = <&gpp16 2 GPIO_ACTIVE_HIGH>;

		vddaon-supply = <&lynx_wcn_buck>;
		vddpmu-supply = <&lynx_wcn_buck>;
		vddpmumx-supply = <&lynx_wcn_buck>;
		vddpmucx-supply = <&lynx_wcn_buck>;
		vddrfa0p95-supply = <&lynx_wcn_buck>;
		vddio-supply = <&lynx_wcn_pmic>;
		vddrfa1p3-supply = <&lynx_wcn_pmic>;
		vddrfa1p9-supply = <&lynx_wcn_pmic>;
		vddpcie1p3-supply = <&lynx_wcn_pmic>;
		vddpcie1p9-supply = <&lynx_wcn_pmic>;

		regulators {
			vreg_pmu_rfa_cmn_0p8: ldo0 { regulator-name = "vreg_pmu_rfa_cmn_0p8"; };
			vreg_pmu_aon_0p8: ldo1 { regulator-name = "vreg_pmu_aon_0p8"; };
			vreg_pmu_wlcx_0p8: ldo2 { regulator-name = "vreg_pmu_wlcx_0p8"; };
			vreg_pmu_wlmx_0p8: ldo3 { regulator-name = "vreg_pmu_wlmx_0p8"; };
			vreg_pmu_btcmx_0p8: ldo4 { regulator-name = "vreg_pmu_btcmx_0p8"; };
			vreg_pmu_pcie_1p8: ldo5 { regulator-name = "vreg_pmu_pcie_1p8"; };
			vreg_pmu_pcie_0p9: ldo6 { regulator-name = "vreg_pmu_pcie_0p9"; };
			vreg_pmu_rfa_0p8: ldo7 { regulator-name = "vreg_pmu_rfa_0p8"; };
			vreg_pmu_rfa_1p2: ldo8 { regulator-name = "vreg_pmu_rfa_1p2"; };
			vreg_pmu_rfa_1p7: ldo9 { regulator-name = "vreg_pmu_rfa_1p7"; };
		};
	};
};

/* &cmu_apm, &usi_apm1 and the bt_reg_on entry under &pinctrl_peric0 as before. */

&serial_apm1 {
	status = "okay";

	bluetooth {
		compatible = "qcom,wcn6855-bt";

		vddrfacmn-supply = <&vreg_pmu_rfa_cmn_0p8>;
		vddaon-supply = <&vreg_pmu_aon_0p8>;
		vddwlcx-supply = <&vreg_pmu_wlcx_0p8>;
		vddwlmx-supply = <&vreg_pmu_wlmx_0p8>;
		vddbtcmx-supply = <&vreg_pmu_btcmx_0p8>;
		vddrfa0p8-supply = <&vreg_pmu_rfa_0p8>;
		vddrfa1p2-supply = <&vreg_pmu_rfa_1p2>;
		vddrfa1p7-supply = <&vreg_pmu_rfa_1p7>;

		/* 3.2 Mbaud garbles every frame on this UART; 1, 2 and 3 Mbaud work. */
		max-speed = <3000000>;
	};
};

The choices, so you can object to any of them:

  1. Two input regulators, not one. qcom,wcn6855-pmu requires ten input supplies. The only one the host switches on lynx is the 0.95 V buck on gph2-5 (stock bt-0p95-en), so the five 0.95 V inputs (vddaon, vddpmu, vddpmumx, vddpmucx, vddrfa0p95; the x13s feeds all five from one 0.95 V rail) point at it. The other five (vddio 1.8 V, vddrfa1p3, vddrfa1p9, vddpcie1p3, vddpcie1p9) are PMIC rails stock keeps on and nothing of ours drives, so they point at a second regulator-fixed with regulator-always-on and no GPIO. That is a stand-in for rails we have not described, and the comment says so. Pointing all ten at the buck would be shorter but would claim the buck feeds 1.8 V. Either is fine with me.
  2. No wlan-enable-gpios. Left out until there is a PCIe node, as you said the binding allows, so the Wi-Fi bench keeps gph2-4 and nothing in its runbooks changes. The message will say which choice was made.
  3. No swctrl-gpios, xo-clk-gpios or clocks. Google's btpower for lynx drives only the buck and BT_EN; I have not found a SW_CTRL or XO line for lynx in its device tree, and the driver treats all three as optional.
  4. The ldo names and numbering follow the x13s (sc8280xp-lenovo-thinkpad-x13s.dts), the published wcn6855 user, so the Bluetooth node's eight supplies read the same as upstream's.

One thing for the module side, which goes in the package, not this PR: hci_qca asks for the bluetooth power sequencer only when the node has no enable-gpios, and if pwrseq_get() fails (including -EPROBE_DEFER when no provider has registered yet) it sets pwrseq = NULL and falls back to driving the regulators itself; with no enable-gpios that means power_ctrl_enabled = false and a controller that never powers up. CONFIG_POWER_SEQUENCING_QCOM_WCN=m, so pwrseq-qcom-wcn has to be loaded before hci_uart. I will put it first in modules-load.d and ship it with the other modules.

Next boot, once you have looked at this and the pin controller is out: the PMU version, with tools/phone-smoke, uname -r, dmesg | grep -i -e qca -e bluetooth -e pwrseq and bluetoothctl show, as you asked.

@FlorencioJP

FlorencioJP commented Oct 29, 2026 •

Copy link
Copy Markdown
Author

@derekross, the evidence boot on the PMU version, tonight, without waiting for your look at the draft, since it only cost a vendor_kernel_boot_b flash (the kernel is the same wifi1 build, the DTB is the draft above, dtc output bd0765b5…).

Boot. uname -r: 7.3.0-rc6-omamobile-lynx-wifi1, slot B. dmesg | grep -i -e qca -e bluetooth -e pwrseq, trimmed to the lines that matter:

[    2.197173] systemd-modules-load[432]: Inserted module 'hci_uart'
[    2.204582] Bluetooth: hci0: setting up wcn6855
[    2.271445] Bluetooth: hci0: QCA SOC Version  :0x400c0210
[    2.287612] Bluetooth: hci0: QCA Downloading qca/wcnhpbtfw21.tlv
[    2.920076] Bluetooth: hci0: QCA Downloading qca/wcnhpnv21.bin
[    3.097351] Bluetooth: hci0: QCA setup on UART is completed
[    3.826179] Bluetooth: MGMT ver 1.23
[    3.826420] Bluetooth: hci0: setting up wcn6855
[    4.698940] Bluetooth: hci0: QCA setup on UART is completed

No line from pwrseq-qcom-wcn itself (it prints nothing on success). That it is the one driving the pins shows in sysfs instead:

  • lsmod: pwrseq_qcom_wcn 16384 1, the one user being hci_uart's serial0-0;
  • /sys/class/regulator: lynx-wcn-0p95-buck enabled, num_users 5 (the five 0.95 V inputs); the always-on stand-in lynx-wcn-pmic-rails has 6 users (five inputs plus regulator-always-on);
  • pinctrl debugfs with Bluetooth on: gpp16-2 (BT_EN) CON(0x1) DAT(0x1), gph2-5 (buck) CON(0x1) DAT(0x1), gph2-4 (WLAN_EN) CON(0x0) DAT(0x0), untouched.

bluetoothctl show: Controller <BT-ADDR> (public), Powered: yes, name pixel-omarchy; a 20 s scan on lists 28 devices. The address is the bootloader's /chosen/config/bt_addr, set by the package's service at 3.82 s.

Power-off. echo serial0-0 > /sys/bus/serial/drivers/hci_uart_qca/unbind: one second later gpp16-2 DAT(0x0), gph2-5 DAT(0x0), lynx-wcn-0p95-buck disabled with 0 users. So the sentence stays: Bluetooth owns gph2-5 and drops it on power-off, on the PMU version as on the fixed-regulator one. The Wi-Fi bench note (unbind first, record which kernel) is unchanged.

Module order. The hazard from the draft is real on this phone: systemd-modules-load inserts hci_uart at 2.19 s with four probe threads, so a plain modules-load.d line cannot order pwrseq-qcom-wcn before it. What works is a modprobe softdep, which libkmod honours on every load path, udev or modules-load:

# /etc/modprobe.d/lynx-bluetooth-pwrseq.conf
softdep hci_uart pre: pwrseq-qcom-wcn

That file and pwrseq-qcom-wcn.ko go into the next omamobile-lynx-bluetooth (0.1.2) with the kernel merge, on the userspace PR.

One thing found on the way, package side, not the DT. After a rebind of serial0-0, the udev-triggered address service runs before hci0 is registered (Set Public Address for hci0 failed with status 0x11 (Invalid Index) at 384.61 s, the controller's setup completes at 385.44 s), so bluez never configures the controller until the script is run again. On a normal boot the order is the other way round (3.10 s setup, 3.82 s address). I'll make the script wait for the index in 0.1.2.

tools/phone-smoke --here, read-only, same boot:

Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-lynx-wifi1, slot b, up 393s
storage guard (required) ok 55 partitions; only userdata and slot B's boot images writable
root (required) ok ext4 on userdata, 1.8G free
kernel health (required) ok no oops or BUG; 0 WARN, 12 error lines in dmesg
systemd units not ok failed: ufw.service; user: xdg-document-portal.service
USB gadget (required) ok bound to 11210000.usb
display ok card0-Unknown-1 connected, 1080x2400
GPU ok render node on panthor
session (required) ok Hyprland and the Omarchy shell running
lock ok PAM file present, the shell can lock
touch ok Goodix Berlin Capacitive TouchScreen
Wi-Fi skip no wireless interface
Bluetooth ok hci0, powered unknown

(ufw.service and the portal failure are the phone's state, not this change; powered unknown is the tool reading D-Bus as root without a session bus.)

Still not run: dtbs_check. If you'd rather change the two input-regulator choices or the ldo naming, the DTS is in my tree and the boot is cheap to repeat.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @FlorencioJP, for sending the draft before spending a boot on it. It's the shape I hoped for. Your reading of the sequencer is right: the buck with its 7 ms startup delay, then BT_EN, then 50 ms, so the order and the delay before BT_EN match the version you tested. Off is BT_EN low and then the regulators, so Bluetooth still drops the buck until a Wi-Fi node takes the wlan target.

The pin controller is published: cb71d6e on omamobile adds gs201's HSI2 pin controller with gph2 and gph4. It's status = "disabled" in gs201.dtsi, and boards that use HSI2 pins enable it. So when you rebase, please add:

&pinctrl_hsi2 {
	status = "okay";
};

Without it the file still compiles and validates, because the gph2 label exists, but at boot the buck regulator's GPIO lookup never resolves. The regulator, the PMU and Bluetooth then wait forever, with nothing in dtc or dtbs_check to say why.

I built your draft on today's omamobile with that block added and the conflict with #93 resolved as before (keep &usi0, append your blocks after it). dtc with W=1 gives no new warnings. dt-validate against omamobile's bindings reports nothing new either: the only message is the B0_IPOP one the base board already has. To make sure the check was live, I removed one required supply from each of the PMU and Bluetooth nodes, and it reported both. CI's dtbs_check will still have the final word after the rebase.

On your four choices:

  1. Two input regulators: fine. The schema accepts a fixed regulator with no voltage, and the sequencer only enables its inputs, so the always-on stand-in works. If Google's lynx device tree names the rails that feed vddio and the 1.3/1.9 V inputs (the supply properties on its cnss2 or btpower node), please name them in the stand-in's comment. If it doesn't, say that in the comment. Not blocking either way.
  2. No wlan-enable-gpios: agreed. Please say so in the message, as you planned.
  3. No swctrl, xo-clk or clocks: fine. The driver treats all three as optional, and with no xo-clk GPIO its assert and deassert steps do nothing except the 50 ms wait.
  4. x13s naming: good, it keeps the Bluetooth node's supplies readable against upstream's.

On the module note, one correction. You're right that hci_qca discards the pwrseq_get() error, -EPROBE_DEFER included, and falls back. But the fallback then asks for its own regulators, and vddrfacmn, vddrfa0p8 and the others point at the PMU's ldo nodes. Nothing registers those nodes as regulators: the sequencer only matches the vddaon-supply phandle. For a node that exists with no regulator behind it, of_regulator_dev_lookup() returns -EPROBE_DEFER. So the probe defers and retries, and gets the sequencer once it's up, rather than ending with a controller that never powers on. What matters is that pwrseq-qcom-wcn is installed: it autoloads from the PMU node's compatible. Listing it first in modules-load.d does no harm. Your pwrseq grep in the boot evidence will show which path ran.

For the rebased commit, the points from before still stand: keep the sentence that Bluetooth owns gph2-5 and drops it on power-off, and run the next boot on the PMU version, with the tools/phone-smoke report, uname -r, dmesg | grep -i -e qca -e bluetooth -e pwrseq and bluetoothctl show. Once CI is green, dtbs_check included, this merges, and omarchy-mobile #98 follows.

Thanks again. Reading the binding and the driver this closely before a boot saves us both time.

@FlorencioJP

FlorencioJP commented Oct 29, 2026 •

Copy link
Copy Markdown
Author

@derekross, rebased onto omamobile at cb71d6e as one commit, 5127dc7: your &pinctrl_hsi2 { status = "okay"; }; block, &usi0 kept and the Bluetooth blocks appended after it, the stand-in's comment says stock's device tree names no supply on its cnss2 or btpower node, and the message states no wlan-enable-gpios and keeps "Bluetooth owns gph2-5 and drops it on power-off". dtc on this exact file against gs201.dtsi at cb71d6e gives only the base's dsi@1c2c0000 warning; dtbs_check is CI's.

Evidence boot on this file's DTB (422214e8… in vendor_kernel_boot_b, kernel still the wifi1 build), uname -r: 7.3.0-rc6-omamobile-lynx-wifi1.

dmesg | grep -i -e qca -e bluetooth -e pwrseq, the lines that matter:

[    2.208537] Bluetooth: HCI UART driver ver 2.3
[    2.209258] Bluetooth: hci0: setting up wcn6855
[    2.278160] Bluetooth: hci0: QCA SOC Version  :0x400c0210
[    2.294155] Bluetooth: hci0: QCA Downloading qca/wcnhpbtfw21.tlv
[    2.926765] Bluetooth: hci0: QCA Downloading qca/wcnhpnv21.bin
[    3.089635] Bluetooth: hci0: QCA setup on UART is completed
[    3.783102] Bluetooth: MGMT ver 1.23
[    3.783362] Bluetooth: hci0: setting up wcn6855
[    4.677275] Bluetooth: hci0: QCA setup on UART is completed
  • lsmod: pwrseq_qcom_wcn 16384 1 (held by hci_uart), so the sequencer path ran, not the fallback;
  • /sys/class/regulator: lynx-wcn-0p95-buck enabled, num_users 5; lynx-wcn-pmic-rails 6 users;
  • pinctrl: gpp16-2 CON(0x1) DAT(0x1), gph2-5 CON(0x1) DAT(0x1), gph2-4 CON(0x0) DAT(0x0);
  • bluetoothctl show: Controller <BT-ADDR> (public), Name: pixel-omarchy, Powered: yes; a 20 s scan lists 22 devices;
  • echo serial0-0 > /sys/bus/serial/drivers/hci_uart_qca/unbind: gpp16-2 and gph2-5 DAT(0x0), the buck disabled with 0 users; bind again: Powered: yes.

tools/phone-smoke --here on that boot: boot, storage guard, root, kernel health (0 WARN, 4 error lines), USB gadget ok; Bluetooth hci0, powered true; Wi-Fi skip. Session, GPU and touch are not ok on this DTB: the board file keeps the GPU disabled, and the touchscreen's ldo25m/ldo26m need the PMIC driver this 25 October kernel binary does not have. Both are the base file's state on this binary, not this change; the local variant we run day to day adds the GPU and the touch, and is where the earlier report with session ok came from. If you want the smoke table on today's omamobile kernel too, say so and I'll build it.

Package side, done on the phone since: omamobile-lynx-bluetooth 0.1.2 ships pwrseq-qcom-wcn.ko, the softdep hci_uart pre: pwrseq-qcom-wcn line (not needed for correctness, as you said; it only makes the order the same on every boot path), and the address script retries while the index is not up. It follows on omarchy-mobile #98 after this merges.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @FlorencioJP. This is the version I hoped for: the PMU node, &pinctrl_hsi2 enabled, the stand-in's comment saying stock names no supply, and a message that says there's no wlan-enable-gpios and that Bluetooth owns gph2-5. The boot evidence on this file's DTB shows the sequencer path ran (pwrseq_qcom_wcn held by hci_uart), the buck with five users, both pins high, and both dropping on unbind.

Our gate on this head, merged onto today's omamobile: it merges cleanly, the sign-off is yours with no AI sign-off, checkpatch has no errors, and gs201-lynx and the other gs201 boards build with W=1 and no new warnings. Nothing outside gs201-lynx.dts changes, so the other gs201 phones are unaffected.

On your question, the smoke table on today's omamobile kernel: no need. The change is device tree only, and the DTB you booted came from this file on cb71d6ef8129. I compared the drivers on the Bluetooth path between the wifi1 tree you ran (1cb4bd36f3be) and this head: pwrseq-qcom-wcn, hci_qca, btqca, the Samsung UART, serdev, the fixed regulator, the gs201 clocks and the USI driver are byte-identical. The Samsung pinctrl differs only in additions for Tensor G3 and G4 that gs201 never takes. So your boot covers what this PR lands. The GPU, touch and session results on that boot are the base board's state on that binary, as you said.

Two small things before I merge:

  1. The file now ends without a newline (\ No newline at end of file after &usi_apm1 { … };). checkpatch warns MISSING_EOF_NEWLINE. Please add it back in the same commit.
  2. A privacy point, about the thread rather than the code. Your comments from 00:46 and 08:57 quote bluetoothctl show with the controller's full address. That's the address the bootloader publishes for your phone, and in this project we keep MAC and Bluetooth addresses out of public posts (<BT-ADDR> is fine). I missed it in the earlier comment, sorry. Could you edit both comments to replace it? I can do it for you if you'd prefer.

After the re-push, CI will need my approval again, and I'll approve it. If dtbs_check comes back green there (our local container can't run it), this merges. omarchy-mobile #98 follows after that. When you update #98, please add the PMU boot to device/lynx/NOTES.md, since the commit message points there for the log excerpts and the notes still describe the fixed-regulator version.

Thanks again for checking the module ordering and the rebind race on the phone. Those notes will save the next person a lot of time.

The Pixel 7a carries the WCN6855 BT core on the APM USI1 UART (ttySAC18
in Google's kernel), as cheetah carries its BCM4389. Enable cmu_apm,
usi_apm1, serial_apm1, pinctrl_peric0 and pinctrl_hsi2, and describe
the controller as qcom,wcn6855-bt powered by a qcom,wcn6855-pmu node
(pwrseq-qcom-wcn): BT_EN on gpp16-2, and the 0.95 V buck shared with
Wi-Fi on gph2-5 as the sequencer's input, a fixed regulator with the
7 ms ramp stock btpower waits before BT_EN. The sequencer then holds
BT_EN high and waits 50 ms. No wlan-enable-gpios is given: WLAN_EN
(gph2-4) stays low until a PCIe node takes the sequencer's wlan target.

With this node Bluetooth owns gph2-5 and drops it on power-off: off is
BT_EN low and then the regulators, so the buck goes down whenever
Bluetooth goes down (stock btpower never drops it), including once at
every boot while the controller has no address yet. The five PMIC rails
behind the chip's I/O, 1.3 V and 1.9 V inputs are an always-on stand-in
regulator: stock's device tree names no supply on its cnss2 or btpower
node, so which PMIC regulators feed them is unknown. No device-wake or
host-wake line is described, so Bluetooth cannot wake the phone.

max-speed is 3 Mbaud: at 3.2 Mbaud every TLV segment of the firmware
download answers with "Frame reassembly failed (-84)" on this UART; 1, 2
and 3 Mbaud work.

Tested on one phone (chip revision 0x400c0210; untested on 0x400c1211)
with hci_uart/btqca, pwrseq-qcom-wcn and linux-firmware's
qca/hpbtfw21.tlv + hpnv21.bin, on 7.3.0-rc6-omamobile-lynx-wifi1 with
this file's DTB: the sequencer holds the buck (five users), BT_EN and
gph2-5 read high, firmware loaded 3 s after boot, controller powered,
a 20 s scan sees the neighbours, and unbinding the serdev driver drops
both pins and disables the buck. The controller reports no public
address; userspace sets the one the bootloader publishes in
/chosen/config/bt_addr. Facts and log excerpts: derekross/omarchy-mobile#98
and device/lynx/NOTES.md there.

These nodes must be enabled in the boot DTB: enabling cmu_apm through a
runtime overlay oopses, since gs201_cmu_probe is __init.

Assisted-by: LLM
Signed-off-by: Jean-Paul Florencio <jpflorencio@version-tracker.app>
@FlorencioJP

Copy link
Copy Markdown
Author

@derekross, both done. The newline is back, in the same commit: a28238c, same parent cb71d6e, same message, only the last byte of gs201-lynx.dts differs from 5127dc7 (my API push had dropped it). Both comments now say <BT-ADDR>; thanks for catching it. I'll add the PMU boot to device/lynx/NOTES.md when I update omarchy-mobile #98.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @FlorencioJP, both are done. a28238c differs from 5127dc7 only in the final newline: same parent, same message, and checkpatch no longer warns. Both comments now read <BT-ADDR>, and I found no other address in the thread, the description or the commit message.

One follow-up on the privacy point, nothing for you to do: GitHub keeps the original text of an edited comment in its edit history, which anyone can open from the "edited" menu, so the address is still visible there for both comments. I've deleted those two original revisions on my side.

Our gate on this head, merged onto today's omamobile: it merges cleanly, the sign-off is yours with no AI sign-off, checkpatch has no errors or warnings, and the gs201 boards build with W=1 and no new device-tree warnings. I also ran dt-validate on gs201-lynx locally against the merged tree's bindings. It reports only the one message the base board already has, and it does flag the Bluetooth node when I break a required supply on purpose, so the check is real. I'll approve the CI run now, and once it's green this merges. omarchy-mobile #98 follows, with the PMU boot in device/lynx/NOTES.md as you said.

Thanks again for the patience across the rounds. This is a clean, well-evidenced node.

@derekross
derekross merged commit 5df2fd3 into derekross:omamobile Nov 3, 2026
1 check passed
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