Skip to content

arm64: dts: exynos: google: gs201-panther: add Wi-Fi - #195

Open
angelrenefm wants to merge 1 commit into
derekross:omamobilefrom
angelrenefm:pr-panther-wifi
Open

angelrenefm wants to merge 1 commit into
derekross:omamobilefrom
angelrenefm:pr-panther-wifi

Conversation

@angelrenefm

@angelrenefm angelrenefm commented Nov 3, 2026 •

Copy link
Copy Markdown
Contributor

Wi-Fi (BCM4389 on PCIe GEN4A_1) for the Pixel 7, as on the Pixel 7 Pro, following derekross's notes on derekross/omarchy-mobile#342. Bluetooth is left to nesmeck's own commit, as agreed on that issue; this one will rebase onto it once it is open.

Why these lines. Stock's PANTHER MP overlay (board id 0x30406, from a Pixel 7's own OTA dtbo) puts WL_REG_ON on gph2-4, WL_DEV_WAKE on gph2-5 and WL_HOST_WAKE on gpa6-1, the same as CHEETAH MP (0x30306), and keeps the PCIe rails (S2MPG13 LDO3S and LDO18S) always on, as cheetah. So the commit takes cheetah's Wi-Fi nodes as they are (the WL_REG_ON regulator, PCIe GEN4A_1 and its PHY, the System MMU's DMA window and the three pin groups), with brcm,board-type = "google,gs201-panther", the name the Pixel 7's firmware files carry.

The restricted DMA pool comes with it: wifi_dma_pool (restricted-dma-pool, 16 MiB below 4 GiB) and the Wi-Fi node's memory-region, as in gs201-cheetah.dts.

Tested on a second Pixel 7, not on omamobile alone. The kernel was 7.3.0-rc6-omamobile-push1-g4446f52aefe8: omamobile at fe9627c, plus #181's series as it stood that day (its five display commits and two zuma UFS commits it carried), plus this commit as first posted, which also carried cheetah's Bluetooth nodes. The Wi-Fi nodes here are the same as in that build.

  • PCIe Gen2 x1 link up on the first try; brcmfmac found BCM4389/3 and loaded the Pixel 7's firmware 20.101.91 (from its own OTA with tools/extract-vendor-firmware -d panther; no firmware in any repository); wlp1s0 came up and a scan saw 33 networks.
  • Then, on the same boot: connected to a 5 GHz network (5500 MHz, signal -55 dBm, rx bitrate 1200.9 Mbit/s) and five pings out had 0 % loss.
  • Kernel messages seen on the next boot (…-g3631214cc15a, the same kernel plus a CMA change): brcmf_cfg80211_add_iface: add iface p2p-dev-wlp1s0 type 10 failed: err=-5 and WLC_SET_ROAM_TRIGGER error (-52). wlp1s0 worked regardless.

Not tested: this commit on omamobile without #181; power use; s2idle with Wi-Fi loaded (the suspend tests unloaded it first, as SUSPEND-PLAN does); this Wi-Fi-only build on the phone (it was built and checked, not booted).

It needs the package side, derekross/omarchy-mobile#419, which blacklists pcie_gs201 and brcmfmac and starts them behind a hang marker (Wi-Fi only too); that one should merge with or before this.

Checks: tools/kernel-pr-check --build OUT origin/omamobile passes everything: the sign-off, no Co-Authored-By, the subject prefix, checkpatch --strict with no warnings, the policy checks, and the 9 device trees with W=1 and dtbs_check (dtschema 2026.9) with no new warnings.

I used AI help (Claude) to work on this.

@angelrenefm angelrenefm changed the title arm64: dts: exynos: google: gs201-panther: add Wi-Fi and Bluetooth arm64: dts: exynos: google: gs201-panther: add Wi-Fi Nov 4, 2026
@nesmeck

nesmeck commented Nov 4, 2026

Copy link
Copy Markdown
Contributor

@angelrenefm, a second run on my Pixel 7, as derekross asked. It's on omamobile (5df2fd3) with only #196 (Bluetooth) and this PR merged on top, so the stacked state, without #181 or anything else of mine. Kernel 7.3.0-rc6-omamobile-push1-gca34bf229d54, with the same omarchy-mobile kernel config as main, and AVB footers as in omarchy-mobile#396. The firmware is this phone's own (tools/extract-vendor-firmware -d panther: brcmfmac4389c1-pcie.google,gs201-panther.{bin,txt,clm_blob}, sha256 checked on the phone). The package side came from omarchy-mobile#419 (identical to cheetah's), installed by hand as bring-up files: the autoload blacklist, and omamobile-wifi-start run from SSH, without the unit or the MAC helper.

Step Result
W1, boot pass: pcie_gs201 and brcmfmac stayed unloaded; 14520000.pcie present; "created restricted DMA pool at 0xff000000, size 16 MiB" for wifi-dma-pool; Bluetooth (hci0, the ES1PX_GG_P10 patch) and the session up, no failed units
W2, omamobile-wifi-start pass: PCIe Gen.2 x1 link up on try 1 (PLL and CDR locked); the endpoint 14e4:4441 "assigned reserved memory node wifi-dma-pool"; firmware BCM4389/3, 20.101.91; wlp1s0 up 2 s after the script started; the hang marker removed
Scan 25 and 26 BSS on two scans (18 at 2.4 GHz, 8 at 5 GHz)
Connect NetworkManager joined a saved network on its own: 5 GHz, -32 dBm, TX bitrate 780 Mbit/s; the router at 0 % loss (2.1 ms on average), the internet at 8.6 ms
W3, Wi-Fi and UFS at once pass, twice. First, for 31 s, HTTPS downloads from the internet over wlp1s0 (360 MB, about 97 Mbit/s, with the server throttling) while 8 rounds of 800 MiB of random data were written to userdata, the page cache dropped, and read back: all 8 sha256 matched. Then over the LAN: a 2 GiB file of random data downloaded to userdata at 618 Mbit/s (27.8 s, from a temporary HTTP server on another machine) while 7 more 800 MiB rounds ran. The download's sha256 matched the server's, and so did all 7 rounds. Neither run left a UFS, SCSI, System MMU, IOMMU, PCIe or brcmfmac error in dmesg, and the link stayed up (TX bitrate 866.7 Mbit/s afterwards). The hottest zone was 68 °C

The log lines worth knowing about, none of which stopped anything:

  • no txcap_blob available (err=-2): the Pixel 7's vendor image has no txcap file;
  • "Ignoring unexpected firmware channel" 169, 173 and 177 (4 times each);
  • WLC_SET_ROAM_TRIGGER error (-52);
  • brcmf_p2p_create_p2pdev: timeout occurred, then "add iface p2p-dev-wlp1s0 type 10 failed: err=-5": no Wi-Fi Direct; station mode is unaffected;
  • two "Received unexpected monitor pkt".

Not tested: the boot unit (omamobile-wifi.service) and the MAC helper, roaming, suspend with Wi-Fi up, power, and a long run.

I'm working on this with AI help (Claude).

🤖 Generated with Claude Code

@angelrenefm

Copy link
Copy Markdown
Contributor Author

@nesmeck, thank you for the stacked run on your Pixel 7, and for the UFS stress alongside it: that's the test this change most needed, and more than mine covered.

I also checked the stacking on my side: this commit applies cleanly on top of #196 (on omamobile 5df2fd3), and tools/kernel-pr-check still passes there, including the 9 device trees with W=1 and dtbs_check. As both merge cleanly in either order, I've left #195 on omamobile rather than carrying a copy of your commit in it; if derekross prefers it stacked, I'll rebase it.

I used AI help (Claude) to work on this.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @angelrenefm, and thank you @nesmeck for the second run. The nodes are cheetah's exactly, with the restricted DMA pool and its memory-region as asked, and the SysMMU left disabled with only its window. Google's own Pixel 7 device tree agrees with your dtbo reading: in kernel/devices/google/pantah (branch android-gs-pantah-6.1-android16), gs201-panther-common.dtsi includes the same gs201-cloudripper-wlan.dtsi as the Pixel 7 Pro, and no panther file overrides those pins.

Checks on my side: tools/kernel-pr-check --build --merge passes in the project container (checkpatch clean, the 9 device trees with no new warnings). I also ran dt-validate on the host: 0 messages before and after, and 0 for this commit together with #196, and a deliberately broken copy was rejected, so the check really ran. The two PRs merge cleanly in either order, so no rebase is needed. CI is waiting for my approval because this is your first PR here; I'll approve the run.

nesmeck's run is the one that matters most for merging: omamobile with only #196 and this commit, link up on the first try, and two runs of Wi-Fi downloads alongside heavy userdata writes with every checksum matching. That's exactly the storage-and-Wi-Fi test this change needed.

Merge order: omarchy-mobile#419 first, so every Pixel 7 has the autoload blacklist before a kernel with this node reaches it, then this one.

Small things, none blocking:

  • The kernel's coding-assistants.rst asks for an Assisted-by: LLM trailer rather than a sentence in the body. If you push again, please swap it in above your Signed-off-by:.
  • The commit message could cite Google's DT files above, and mention nesmeck's stacked run as the test on omamobile.
  • wlan_host_wake could go into the board's existing &pinctrl_far_alive block instead of a second one.

Thanks for a careful, well-documented first kernel PR.

The Pixel 7 has the Pixel 7 Pro's BCM4389 on the same PCIe lines: stock's
panther MP overlay (board id 0x30406) puts WL_REG_ON on gph2-4,
WL_DEV_WAKE on gph2-5 and WL_HOST_WAKE on gpa6-1, as cheetah's does, and
keeps the PCIe rails always on. Google's own device tree agrees: in
kernel/devices/google/pantah (branch android-gs-pantah-6.1-android16),
gs201-panther-common.dtsi includes the same gs201-cloudripper-wlan.dtsi as
the Pixel 7 Pro, and no panther file overrides those pins. Take cheetah's
Wi-Fi nodes: the WL_REG_ON regulator, the restricted DMA pool and the
Wi-Fi node's memory-region, PCIe GEN4A_1 and its PHY, the System MMU's DMA
window and the pins, with the board type the Pixel 7's firmware names
carry. Bluetooth is left to its own commit.

On a Pixel 7, a kernel with these nodes and the Bluetooth ones brought up
the link at Gen2 x1, found BCM4389/3 with the phone's own firmware, and
connected wlp1s0 to a network. A second Pixel 7 ran omamobile with only
the Bluetooth commit and this one: the link came up on the first try, and
in two runs of Wi-Fi downloads alongside heavy writes to userdata every
checksum matched.

Assisted-by: LLM
Signed-off-by: Ángel René <17342133+angelrenefm@users.noreply.github.com>
@angelrenefm

Copy link
Copy Markdown
Contributor Author

@derekross, thanks for the review. I've pushed the three small things, still as one commit:

  • wlan_host_wake now sits in the board's existing &pinctrl_far_alive block, next to bt_host_wake and hapt_amp_irq, instead of a second block.
  • The commit message cites Google's gs201-panther-common.dtsi including gs201-cloudripper-wlan.dtsi (kernel/devices/google/pantah, android-gs-pantah-6.1-android16), and mentions @nesmeck's stacked run on omamobile as the test.
  • The AI sentence is gone from the body; there's an Assisted-by: LLM trailer above my Signed-off-by:.

Since #196 and the flash LED commit have landed, I rebased onto the current omamobile (29a7a93). The only conflict was both commits adding nodes at the end of gs201-panther.dts; I kept both.

Moving the pin changes nothing in the build: gs201-panther.dtb from the rebased commit before the move and after it are byte-identical, both plain and with -@ (so __symbols__ too), and dtc -I dtb -O dts shows no difference.

Checks: tools/kernel-pr-check --build --merge omamobile passes: it merges cleanly, the sign-off is mine, there are no Co-Authored-By trailers, checkpatch --strict gives 0 errors and 0 warnings, the policy checks pass, and the 10 device trees build with W=1 and dtbs_check with no new warnings. dt-validate gives 0 messages on gs201-panther.dtb before and after the change. A copy with a made-up property in wlan-host-wake-pins was rejected, so the check really ran.

I used AI help (Claude) to work on this.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @angelrenefm. All three small points are in, and the rebase onto today's omamobile looks right: comparing your new commit with the one I reviewed shows only the message and the moved pin group. tools/kernel-pr-check --build --merge passes in the project container, and the device tree validates with no messages.

One thing to know: this and your fingerprint PR (#207) both add their nodes at the very end of gs201-panther.dts, after the flash LED, so whichever lands second will conflict. Keeping both blocks is the whole resolution. I built the tree with this one, the fingerprint one and reboot=warm together that way and it builds and validates cleanly, so I can resolve it at merge if you'd rather not rebase again.

Order stays as before: omarchy-mobile#419 first, published, then this one.

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.

3 participants