notes/28 — 2026-06-20 — B-USB bring-up (USB2PHY + DWC2)

Status: code complete + built-in; board validation PENDING. The only active peripheral left. Goal = USB enumeration (dmesg device descriptor); WiFi RTL8733BU dongle driver is follow-up.

The three gaps (all wiring + one driver of_match, no architecture 原创)

1. Driver: mainline drivers/phy/rockchip/phy-rockchip-inno-usb2.c had no rk3506-usb2phy of_match → PHY never probed. Vendor 6.1 fork supports it but on a diverged model (extra phy_base mmio + extra cfg fields + macros).
2. DT: forge rk3506.dtsi (minimal) had no usb2phy/dwc2 nodes.
3. config: PHY_ROCKCHIP_INNO_USB2 off; USB_SUPPORT cut by kernel-trim.

Key architecture finding (the de-risk — read before touching this driver)

Line-by-lined the vendor 6.1 driver. RK3506's USB2PHY uses two register spaces:

• Detection / power / charger (phy_sus@0x0060, bvalid/idfall/idrise/ls_det@0x0150, utmi_*@0x0118, chg_det) → all via rphy->grf (vendor DT rockchip,usbgrf = <&grf>, GRF@ff288000). Mainline's grf-regmap path covers these with zero driver change.
• Analog tuning + 480M clkout gate (clkout_ctl_phy@0x041c, eye-height, pre-emphasis, diff-rx) → via rphy->phy_base (ioremap PHY block@ff2b0000). Mainline struct has no phy_base / no ioremap.
• Why host enumeration doesn't need the hard parts: phy_sus (the actual power switch) is in the GRF → mainline can power the PHY. The vendor DWC2 DT consumes only CRU clocks (HCLK_USBOTGx/_PMU/_ADP), not the PHY 480M → the clkout gate is irrelevant. Analog tuning is signal-quality, not enumeration.

Approach: faithful-lite (not vendor-verbatim, not bare grf-only)

• struct rockchip_usb2phy: +1 field void __iomem *phy_base.
• probe: devm_platform_ioremap_resource(pdev, 0) → phy_base, IS_ERR→NULL (classic GRF-only PHYs unaffected; only RK3506's node has the MEM resource).
• rk3506_usb2phy_tuning(): ports vendor's 6 phy_base writes, open-codes phy_clear_bits/phy_update_bits as readl/writel RMW; no-op if phy_base NULL.
• rk3506_phy_cfgs[]: 2 ports, detection regs in grf (mapped utmi_iddig→utmi_id). Dropped: OTG iddig-force fields (bvalid_grf_sel/con, iddig_output/en, vbus_det_en, port_ls_filter_con) + chg_det (chg_mode has no mainline struct member; charger detect not needed for host) → zero struct field additions. No .clkout_ctl (DT carries no #clock-cells → clk480m_register never runs).
• of_match: +rockchip,rk3506-usb2phy.

Deliverables (patches 0014 + 0015, committed in explore/linux)

• 0014 phy: rockchip: inno-usb2: add RK3506 support (539b4241d)
• 0015 ARM: dts: rockchip: rk3506b-aes: add USB2PHY + DWC2 (75d13ceca)
  • rk3506.dtsi: usb2phy@ff2b0000 (rk3506-usb2phy, usbgrf=<&grf>, no #clock-cells)
    • u2phy_otg0(otg-port,IRQ75/76/77) + u2phy_otg1(host-port,IRQ80/81/82); usb@ff740000 + usb@ff780000 (rockchip,rk3066-usb/snps,dwc2 fallback — no dwc2 driver patch, dwc2 binds on the generic fallback).
  • rk3506b-aes.dts: both PHY ports + both controllers dr_mode="host".
• config: board/rk3506-evb/kernel.config += PHY_ROCKCHIP_INNO_USB2=y + USB_DWC2_HOST=y; kernel-trim.config dropped the # CONFIG_USB_SUPPORT is not set line (it's applied AFTER kernel.config and was gating all USB off — the build first came up with DWC2/PHY absent until this was fixed).
• series: 0014/0015 appended.
• Image: update-usb-bringup.img (34341450 B, md5 fd3e3912c26ec030e42274629f11cf76) in /mnt/d/DownloadFromInternet/. PROVISION-UBIPROG variant (boot.img with the new kernel, padded to fill the 16 MiB boot partition).

Build verification (local, all green)

• --just-dtb: USB nodes compile, all label/clock refs resolve.
• full build: exit 0. .config confirms USB_SUPPORT=y / USB_DWC2=y / USB_DWC2_HOST=y / PHY_ROCKCHIP_INNO_USB2=y / USB_STORAGE=y; modules.builtin lists phy-rockchip-inno-usb2 / dwc2 / usb-storage / usb-common (built-in).

Board test #1 (boot-sdl-202606200858) — FAILED, root cause found + fixed

Boot itself fine (kernel 7.1.0, AES board, ubiprog provisioned rootfs, buildroot login). But USB dead:

[   12.010085] platform ff740000.usb: deferred probe pending: dwc2: error getting phy
[   12.010816] platform ff780000.usb: deferred probe pending: dwc2: error getting phy

and dmesg|grep showed ONLY those two lines — no rockchip-usb2phy ff2b0000 probe line at all. So the PHY provider never came up; DWC2 correctly deferred.

Root cause (DT bug, not driver): the SoC usb2phy parent node is status="disabled" and I'd only enabled the child ports (&u2phy_otg0/1). inno-usb2 binds to the parent (the node with the rk3506-usb2phy compatible); the otg-port/host-port children are the phys it publishes AFTER it probes. With the parent disabled, OF core never creates its platform device → driver never probes → no phy provider → DWC2 defers forever. Driver patch + DWC2 are fine.

Fix (amended into 0015): enable the parent in aes.dts — &usb2phy { status = "okay"; };. Mirrors vendor rk3506-evb2-v10.dtsi:779 (which enables &usb2phy alongside &u2phy_otg0 + &usb20_otg0). New image: md5 c633f11a894d7384f06ca941b7d8eb7d (was fd3e3912).

Board test #2 (boot-sdl-202606200858, after the parent-enable fix) — PASSED ✅

Same image family, &usb2phy { status="okay"; } added. Full success — T1+T2+T3:

dwc2 ff740000.usb: DWC OTG Controller / new USB bus registered, bus number 1
dwc2 ff780000.usb: DWC OTG Controller / new USB bus registered, bus number 2
usb 2-1: new high-speed USB device number 2 using dwc2          ← enumeration
usb 2-1.3: new high-speed USB device number 3 using dwc2         ← behind a hub
usb-storage 2-1.1:1.0: USB Mass Storage device detected
sd 0:0:0:0: [sda] Attached SCSI removable disk                   ← /dev/sda, block device

Both DWC2 controllers registered as host (bus 1+2), got irqs 47/48, got their PHY (no more "error getting phy"). A USB hub + mass-storage stick enumerated and the stick attached as /dev/sda. supply vusb_d/vusb_a not found → dummy regulator is the R2 vbus non-issue, confirmed harmless. B-USB closed. The WiFi dongle is presumably one of the other enumerated devices — driver bring-up is the next phase.

Board-test protocol (user: Windows RKDevTool烧 + UART抓log)

Boot args at the => prompt: console=ttyS0,1500000 ONLY (provision variant — no root=, or /init won't run). Then mtd read boot 0x04000000 0 0x1000000; bootm 0x04000000.

• T1 probe: dmesg | grep -iE "usb2phy|dwc2|u2phy|ff2b0000|ff740000|ff780000|rockchip-usb2phy" → expect rockchip-usb2phy ff2b0000.usb2-phy probed + dwc2 ff740000.usb/ff780000 host controller registered.
• T2 enumerate: plug USB dongle (or USB-stick) → dmesg | grep -iE "usb [0-9]-[0-9]|new (high|full)-speed|device descriptor|Product:" → device descriptor read = B-USB achieved.
• T3 (if USB-stick): ls /dev/sd* / mount.

Risks / follow-up

• R1 PHY analog power-on: vendor tuning touches no SIDDQ (analog block presumed powered by default), so enumeration should work on the first flash. If silent, add a phy_base power-on sequence (mirror whatever vendor's power_on does in phy_base).
• R2 vbus: no vbus regulator wired (board USB-A 5V assumed always-on). If a device doesn't enumerate, add a fixed always-on vcc5v0_host + vbus-supply.
• F1 WiFi RTL8733BU dongle driver (rtw88/rtl8xxxu, mainline, needs config) — the actual goal past enumeration.
• F2 OTG/iddog + gadget: needs the dropped iddig-force cfg fields + extcon.
• F3 480M clkout: only if a later EHCI/OHCI host controller consumes the PHY clock.