NFC

NFC works with Harmattan's own stack: the kernel carries Nokia's PN544 character-device driver (/dev/pn544), which is what Harmattan's nfcd needs. Switching NFC on and off from Settings, and detecting and reading a target (an Android phone), work. Reading an NDEF tag's content, and NFC during CORE OFF, haven't been tested yet.

What Harmattan expects

Mainline 6.18 has only the NFC-subsystem PN544 driver, which Harmattan can't use. Stock had Nokia's drivers/misc/pn544.c (pn544.ko):

• One user. Only libNFCDAL opens the device, for nfcd (user nfc, D-Bus activated as com.nokia.nfc; no upstart job). It opens it with O_RDWR | O_NONBLOCK and watches it with poll(). Each read()/write() carries one LLC frame; the HAL builds the frames and CRCs, and the driver checks the CRC on write.
• One ioctl, PN544_SET_FW_MODE (0x40045002). Chip power follows open()/close().
• udev: /etc/udev/rules.d/92-n770.rules gives the device to nfc, so it shows as crw-rw---- nfc root.
• On/off is an mce radio state (NFCRadioState, off by default). nfcd starts at every boot anyway (conndlgs' NFC agent asks for it), checks the chip's firmware, and closes the device again while NFC is off.

Without the driver, nfcd failed and was restarted about 300 times a minute for the first minutes of every boot, with conndlgs logging "RegisterUIAgent ... com.nokia.nfc.Adapter doesn't exist".

The driver

drivers/misc/nokia-pn544.c, commit 34c9fae34 misc: add Nokia's PN544 NFC driver (/dev/pn544) for the N9, built in (CONFIG_NOKIA_PN544=y). It is stock's code, with the board file's callbacks replaced by DT GPIOs and a regulator:

Bus	I2C3 at 400 kHz, address 0x2b (DT node nfc@2b in omap3-n9-nokia.dtsi)
Interrupt	GPIO 76, rising edge; its pad has the IO wake bit, as NOLO sets it
Enable	GPIO 77
Firmware reset	GPIO 78
Supplies	Vdd_IO = VIO; VBat is the battery directly

NOLO leaves the three pads as GPIOs, so the driver does no pinmuxing. At probe it runs stock's self-test: it powers the chip for about 50 ms, sends a U-RSET link frame and expects 03 e6 17 a7 back. This only talks to the chip and writes nothing to its memory. If the chip doesn't answer, no /dev/pn544 is created.

Differences from stock, on purpose

1. Firmware download mode is refused (-EPERM, kernel log "firmware download mode refused (allow_fw_mode=0)") unless nokia_pn544.allow_fw_mode=1 is set, either on the kernel command line or at /sys/module/nokia_pn544/parameters/allow_fw_mode. nfcd compares the chip's four version bytes with the firmware image built into libNXP_NFC_HAL.so.1.0.0, and on any mismatch rewrites the chip's flash. On a phone that ran stock nfcd, the two match and nothing is ever refused. If the message does appear, the chip's firmware differs from Harmattan's, and allowing it means a flash write on the NFC chip. The HAL's normal configuration writes at each NFC start are not affected, as on stock.
2. The nfc_test sysfs file is left out. It was the board's GPIO self-test, read only by stserver in factory test mode at a 2.6.32 path, and stock's version left the chip powered in firmware mode.
3. The ioctls and open/close run under a mutex: stock relied on the big kernel lock.

Removed and switched off lists the first two, with how to put each back.

Checking it

As root on the phone:

dmesg | grep -i pn544                 # no "rset" error, no "firmware download mode refused"
ls -l /dev/pn544                      # crw-rw---- nfc root
grep pn544 /proc/interrupts           # counts only while NFC is on (and once at probe)
grep -E 'gpio-7[78] ' /sys/kernel/debug/gpio   # "enable" and "firmware": low while NFC is off

NFC on and off, as Settings does (mce's radio state bit 16):

dbus-send --system --type=method_call --dest=com.nokia.mce /com/nokia/mce/request \
  com.nokia.mce.request.req_radio_states_change uint32:16 uint32:16   # on; uint32:0 uint32:16 for off

With NFC on, nfcd opens the device, ENABLE goes high and FW stays low, and the chip interrupts during HCI start-up. With NFC off, the device is closed, both lines go low, the interrupts stop, and nfcd stays running, as on stock.

Tested

• The probe self-test passes; udev gives /dev/pn544 to nfc.
• nfcd starts once at boot, checks the chip's firmware (it matches, so there is no firmware-mode request), and closes the device. The restart loop and the conndlgs errors are gone.
• With NFC on, an Android phone held to the N9's back was detected and read twice through Harmattan's stack. It reported "Tag doesn't contain NDEF data", which is expected from a phone presenting itself as a card.
• NFC off closes the device and powers the chip down.

Not done

• Reading an NDEF tag's content (a sticker, or an Android app that emulates an NDEF tag), and Bluetooth pairing or sharing through NFC.
• NFC on during CORE OFF. GPIO 76's pad wakeup is in the DT (rm696 aed6f173d, every pad as stock runs it). With NFC off the chip is unpowered and its interrupt quiet, so idle shouldn't change.