[AS7327-56X] Upgrade kernel from 4.14 to 6.12

[vincentchiang-ec]

Summary

Brings the AS7327-56X platform onto kernel 6.12 LTS. The platform sources were carried over from support_new_platform_and_sync (which still targeted 4.14) and ported in-place; nothing else on this branch moves.

Config

• modules/PKG.yml, modules/builds/Makefile: KERNELS switched to onl-kernel-6.12-lts-x86-64-all
• platform-config/r0/.../x86-64-accton-as7327-56x-r0.yml: kernel anchor *kernel-4-14 → *kernel-6-12

Build layout

• Moved every .c from modules/builds/ into modules/builds/src/ with an explicit Kbuild (obj-m += ...). tools/scripts/kmodbuild.sh builds each top-level .c in its own mktemp -d, alone with obj-m := <single>.o, so cross-module EXPORT_SYMBOLs (cpld_read/write consumed by leds/psu/psu_bmc, fpga_pcie_read/write, at24_read_eeprom) fail modpost. The src/ layout puts the whole tree in one tmp dir so Module.symvers is shared. Same convention as as5835-54x / as7816-64x.

Driver API (4.14 → 6.12)

• cpld.c, psu.c, at24_as7327_56x.c: i2c_driver.probe() is single-arg; pull the device id via i2c_client_get_device_id(client) where the body still needs id->driver_data.
• cpld.c, psu.c, at24_as7327_56x.c: i2c_driver.remove() returns void.
• leds.c, psu_bmc.c, thermal_bmc.c: platform_driver.remove() returns void.
• psu.c: hwmon_device_register_with_info(chip=NULL) returns -EINVAL on 6.12. Add a stub hwmon_chip_info (is_visible → 0, HWMON_C_REGISTER_TZ) so the real attrs still come from the existing attribute_group.
• thermal_bmc.c: same NULL-chip problem; switched to hwmon_device_register_with_groups() which is a better fit for the sysfs-only use here.
• fpga.c:
  • class_create() lost its owner argument in 6.4.
  • access_ok() lost VERIFY_READ/WRITE in 5.0; also cast the bare unsigned long ioctl arg to (const void __user *) for the stricter prototype check.
  • PCI error handler argument: enum pci_channel_state is no longer a valid enum tag in 6.12; use pci_channel_state_t.
  • I2C_CLASS_SPD was removed; drop it from the adapter class mask.
• at24_as7327_56x.c:
  • i2c_new_dummy() was removed in 5.7; switch to i2c_new_dummy_device() (returns ERR_PTR — error-cleanup uses IS_ERR_OR_NULL).
  • <linux/platform_data/at24.h> was removed in 5.5; inline the small at24_platform_data struct + AT24_FLAG_* macros this driver actually consumes.

IDPROM ↔ kernel SPD-scan collision

Kernel 6.12 i2c-i801 now calls i2c_register_spd() at probe time, which walks DMI memory-device records and, for every populated DDR4 slot, registers a kernel-owned i2c_client named ee1004 at 0x50 + slot on the i801 SMBus. On AS7327-56X the IDPROM EEPROM is wired to i2c-0 0x50, so the SPD scan claims the slot first; the userspace new_i2c_device('accton_24c64', 0x50, 0) then silently fails (address busy) and the device stays unbound — no eeprom sysfs file, onlpdump shows blank System Information. The kernel-owned client is not in adap->userspace_clients, so delete_device cannot remove it (delete_device: Can't find device in list).

Fix: add an ee1004 alias to at24_as7327_56x's id_table, mapped to the same AT24_DEVICE_MAGIC(65536/8, AT24_FLAG_ADDR16) flags as accton_24c64. The platform driver then binds to the pre-existing kernel-instantiated client, nvmem exposes the standard eeprom file, and ONIE TLV decode works as usual. Safe to keep platform-local because at24_as7327_56x.ko only auto-loads on this platform.

Verification (DUT, fresh onie-nos-install)

• uname -r → 6.12.73-OpenNetworkLinux
• onlpdump System Information decodes the full TLV: Product 7327-56X-O-AC-F, Serial 732756X2545006, MAC 94:ef:97:a5:2c:c0, Vendor Accton, ONIE Version, Manufacture date.
• thermal × 10 PRESENT (PSU-1 thermal-{1,2,3} correctly FAILED while PSU-1 is unplugged).
• led × 4 PRESENT.
• psu @ 1: PRESENT,UNPLUGGED, caps=0 (no AC). psu @ 2: full readings, Vin=227V / Vout=12V / Pout=106W.
• fan × 8 chassis PRESENT,F2B; PSU-1 fan PRESENT,FAILED.
• SFP probe matches plugged ports (0, 7, 49); port 49 QSFP shows TX_DISABLE.