Skip to content

Skycontroller 2 system volume: hardware and firmware blobs (static analysis)

Static read of the live Skycontroller 2’s system volume (mpp-linux-1.0.7), hardware and firmware blobs, by one agent, 2026-10-04. Status: Agent-cited, with these claims re-checked by the commissioning session against the copy: etc/boxinit.rc product block (model Skycontroller 2, id 0x090f, updater mpp_update.plf, hostname mpp, IMU icm20608, magnetometer ak8963); the bottom-board hex files parsed by me (18,656 and 18,624 bytes, device descriptor 12 01 00 02 00 00 00 08 eb 03 06 24 00 01 01 02 00 01 in both, meaning USB 2.00, VID 03eb, PID 2406, one configuration, serial string index 0, and the product and manufacturer strings stored as ASCII near the descriptor); and that libpuf.so exports puf_get_version, puf_get_app_id, puf_get_target_id and puf_compare_version, consistent with an update-package reader. Not re-derived: the HID report layout (22 buttons and 5 axes), the Wi-Fi driver and firmware versions, the module inventory, the absence claims (each was checked by the agent against the nap tree as a control), and the iAP2 reading of g_aea.ko, acid.so and iosd. The kernel, pin map and SoC revision are not in this copy.

Do not republish vendor firmware contents. The MAC-shaped strings in the Wi-Fi blobs are vendor defaults and are not quoted here.

Subject: the extracted system volume (bin sbin lib usr etc), build mpp-linux-1.0.7 (original uid mpp-linux-1.0.7-rc4), ro.build.date Tue Oct 24 09:10:16 UTC 2017 (epoch 1508836216 checks out). All paths below are relative to that root unless marked nap: (the earlier image at firmware/extracted/skycontroller-nap/rootfs). Nothing was connected to. recon-1.txt was not opened. No MAC, serial, key or device id is quoted: the Wi-Fi nvram file contains a macaddr= line and I printed the file with that line redacted.

Tags: V = Verified (bytes or listing read), A = Agent-cited (relayed by the caller or an earlier note), I = Inference, O = Open, R = Recollection (outside knowledge, with what would confirm it).

Scratch artifacts (scratchpad, not the repo): bottomboard.hex.bin and bottomboard.dv.hex.bin (flat images of the two hex files, address 0 upward).

  1. The image names itself: ro.product.model "Skycontroller 2", ro.product.model.id "0x090f", ro.updater.filename "mpp_update.plf" (etc/boxinit.rc lines 36 to 44). This is the first evidence in either image that ties 0x090f to the marketing name. (V)
  2. Bottom board is an AVR XMEGA flashed over Atmel DFU with dfu-programmer atxmega32c4. Two hex files: bottomboard.hex (18,656 bytes of code, used when ro.revision is above 4) and bottomboard.dv.hex (18,624 bytes, “legacy (DV)” firmware for revision 4 and below). Both present USB 03eb:2406 “MPP GAMEPAD” with identical descriptors. (V)
  3. HID report 1 is 22 buttons plus 5 signed 8 bit axes (-100 to 100); report 2 is a 16 byte vendor channel each way. The nap board had 24 buttons and 8 axes. The 5 axes line up with mppd’s AXIS_LEFT_LR/UD, AXIS_RIGHT_LR/UD, AXIS_SLIDER_LR. (V)
  4. Radio is one Broadcom BCM43526 on USB, STA only (wifid --mode STA, driver bcmdriver, module set bcm43526-extra). Firmware bcm43526-firmware.bin and driver both report version 6.37.114.64 (r516176); the -extra wl.ko differs from the plain one by channel-interference (chanim) and packet-delay iovars. (V)
  5. SoC is a Parrot7 with a P7MU power unit, by evidence in this image: Parrot P7 drivers in modules.builtin, p7-temperature P7 and P7MU temperature readers in libmpphw.so, console ttyPA3, manifest kernel tree linux-p7. Exact revision is not in the volume. (V for the evidence, I for the conclusion)
  6. Absent in this build: display and framebuffer, HDMI, GPU and video decoder, GPS (mppd’s own hardware hook returns 0), Bluetooth, a Wi-Fi AP (hook returns 0), 5 GHz (hook returns 1, 2.4 GHz only by inference). Present: IMU icm20608 and magnetometer ak8963 (names only; drivers are built into the kernel), an iPod/iAP USB gadget with a coprocessor on I2C, LEDs and battery handled by the bottom board. Every negative was checked against the nap tree as control. (V, I)
  7. libpuf.so is not a physically unclonable function: it parses .plf and tar.gz update packages (app id, target id, version). acid.so is an iPod/iAP accessory stack. pal-*.so are the Parrot platform abstraction library. (V)
  8. This copy has no kernel image, no pin map and no machine descriptor, so GPIO numbers for RESET_WIFI, RESET_USB_HUB, POWER_KEEP, the SoC revision and the board name string must come from the kernel partition. See the last two sections.

usr/bin/bottom_board_update (shell script): reads the DFU bootloader version with dfu-programmer atxmega32c4 get bootloader-version into property ro.bb.bootloader.version; reads gprop ro.revision 0; if the value is greater than 4 it flashes /lib/firmware/bottomboard.hex, otherwise it logs “Flashing legacy (DV) firmware” and flashes /lib/firmware/bottomboard.dv.hex; both with --force --validate-first --ignore-outside. It then writes 1 to user_gpio/POWER_KEEP/value and runs dfu-programmer ... launch. lib/udev/rules.d/52-bottomboard.rules runs the updater when USB 03eb:2fd9 (Atmel DFU) appears, and runs usr/bin/gamepad_plugged when 03eb:2406 appears; that script writes 0 to POWER_KEEP. So POWER_KEEP is a power-hold line that the updater asserts while the board is in DFU and releases once the running firmware enumerates. (V for the scripts, I for the purpose)

ro.revision source: sbin/boxinit strings carry the pair table ro.boot.serialno/ro.serialno, ro.boot.hardware/ro.hardware, ro.boot.revision/ro.revision, plus a /proc/cmdline parse into ro.boot.%s. So the board revision very likely arrives on the kernel command line. (V strings, I for the flow). This replaces the nap kernel’s /sys/kernel/hsis/pcbrev scheme: etc/boxinit.rc here contains no hsis reference (/usr/bin/grep -rn hsis etc returns nothing; control: the same command finds ro.revision in etc/boxinit.d/98-iosd.rc). etc/boxinit.d/98-iosd.rc says that on ro.revision=0 iosd is stopped “due to the usb0 bug”. So revisions 0 up to at least 5 exist. (V)

Both files: Intel HEX, record types 0 and 1 only, one contiguous run from 0x0000, no gaps, no extended address records.

bottomboard.hex bottomboard.dv.hex
text size 52,496 52,406
data 18,656 bytes, 0x0000 to 0x48df 18,624 bytes, 0x0000 to 0x48bf
code family AVR: first 0x1fc bytes are 127 vector slots of 4 bytes, mostly RJMP + nop (xx c0 00 00), vector 10 (offset 0x28) is a JMP (0c 94) same layout

Family: AVR XMEGA. The vector layout is AVR; “XMEGA” and “32C4” come from the updater script naming atxmega32c4 and the bundled dfu-programmer carrying that target (string present, name at file offset 0xfef4 in usr/bin/dfu-programmer). 18.6 KB fits the 32 KB application flash of that part (nap decode: row flash size 0x8000, A/V from the nap report). I did not disassemble AVR code beyond the vector table and diffs. (V, I)

Offsets are in the flat image of bottomboard.hex; the dv file is the same minus 0x28.

  • Device descriptor at 0x4854 (dv 0x482c): 12 01 00 02 00 00 00 08 eb 03 06 24 00 01 01 02 00 01: USB 2.00, class 0 per interface, ep0 8 bytes, VID 0x03eb, PID 0x2406, bcdDevice 0x0100, iManufacturer 1, iProduct 2, iSerial 0 (no serial string), one configuration.
  • Configuration descriptor at 0x482c (dv 0x4804): total length 41, one interface, bus powered with remote wakeup, bMaxPower 100 mA (value 0x32 times 2 mA). Interface 0: HID class 3, subclass 1 (boot), protocol 3, two endpoints, HID descriptor with report descriptor length 0x50 (80 bytes). Endpoints: IN 0x81 interrupt, OUT 0x02 interrupt, both 32 bytes max packet, interval 10 ms.
  • Strings stored as ASCII in flash and widened at send time: MPP GAMEPAD at 0x487c, www.parrot.com at 0x4888, language id table 04 03 09 04 (0x0409) at 0x4896. The relayed live enumeration “www.parrot.com MPP GAMEPAD” gives the order manufacturer then product; so I read manufacturer = www.parrot.com, product = MPP GAMEPAD. (A for the live order, I for the mapping)
  • The 03eb VID is Atmel’s, with Parrot’s PID 0x2406 inside it. usr/share/usb.ids lists 03eb:2406 as “MP3 Player”, a stale unrelated entry; the udev rules treat 03eb:2406 as mpp_gamepad and mpp_bb (lib/udev/rules.d/55-juba-gamepad.rules, 56-juba-bb.rules). (V)

HID report descriptor at 0x47bd (dv 0x4795), 80 bytes, identical in both files (V)

Section titled “HID report descriptor at 0x47bd (dv 0x4795), 80 bytes, identical in both files (V)”

Decoded:

  • Report ID 1, Generic Desktop Joystick (usage 4), collection application.
    • Buttons: usage page 9, usages 1 to 22, 22 inputs of 1 bit, then 2 bits of constant padding (24 bits total).
    • Axes: usage page 1, usages 0x30 to 0x34 (X, Y, Z, Rx, Ry), logical -100 to +100 (15 9c 25 64), 8 bits each, 5 axes.
    • Total input report: ID byte plus 3 bytes of buttons plus 5 bytes of axes = 9 bytes.
  • Report ID 2, vendor page 0xFF00, usage 1, 16 byte input and 16 byte output (logical 0 to 255). This is the hidraw channel (mpp_bb) that libmppbb.so uses.

What libmppbb.so does with report 2 is visible only as exported function names: mpp_bb_led_set, mpp_bb_led_blink, mpp_bb_axis_threshold_set, mpp_bb_stay_on_set, mpp_bb_rtc_timestamp_set, mpp_bb_auto_reboot_set, mpp_bb_set_zero_calibration(_str), mpp_bb_set_min_max_calibration(_str), mpp_bb_set_raw_mode, mpp_bb_battery_state_str. So the bottom board carries LEDs (colored, mpp_bb_cli on <color>), an RTC timestamp, battery state, stay-on and auto-reboot control, and stick calibration. I did not decode the frame layout (O).

Software axis names in usr/bin/mppd and usr/lib/libmppd_hw_mpp.so strings: AXIS_LEFT_LR, AXIS_LEFT_UD, AXIS_RIGHT_LR, AXIS_RIGHT_UD, AXIS_SLIDER_LR, and buttons BUTTON_A B C X Y START LEFT_JS RIGHT_JS plus L1, R1, L2, R2 in etc/mppd/mapping_bebop_2.cfg comments. Five axes in the descriptor match five software axes; the slider is probably the camera tilt control (I). (V strings, I mapping)

How the two files differ (V for the diff, I for the reading)

Section titled “How the two files differ (V for the diff, I for the reading)”
  • Same USB device, configuration and HID descriptors, same vector layout. Byte for byte the descriptor tail through the 80 byte report descriptor is equal; the only differences after it are code addresses embedded in the following data.
  • A binary diff gives 967 differing opcode ranges. Most are address operands from code that moved (a 0x148 byte interrupt routine and a 0x12 byte helper sit at different places in the two files) and from a different SRAM variable layout: the dv build addresses its state variables in a block that is laid out differently, with the main build holding more variables (the dv file is 32 bytes smaller overall). The main build also contains a 26 byte compare-style routine near 0x4750 that the dv build lacks.
  • No version string, build date or tag is present in either file (strings -a over the flat image gives only the two USB strings), so I cannot give either firmware a version. The manifest pins the source package drones/mpp/atmel (packages/atmel-mpp) at revision de2776038c..., which is also the revision listed in the nap image’s manifest; this is a source tree state, not a firmware version. (V)
  • Why “DV”: the script says “legacy (DV)” for revisions 4 and below. R: DV is usually “design validation”, the hardware build stage after engineering samples and before production validation; what would confirm it is a note in the atmel-mpp source or a revision table from Parrot. Not in this volume.
sc2 (mpp) nap
file lib/firmware/bottomboard.hex and .dv.hex lib/firmware/xmega.hex
size 18,656 and 18,624 bytes 13,122 bytes
USB 03eb:2406 “MPP GAMEPAD” 03eb:2407 “NAP GAMEPAD”
buttons 22 24
axes 5, -100 to 100, 8 bit 8, -100 to 100, 8 bit
vendor report 16 bytes each way 16 bytes each way
flash command dfu-programmer atxmega32c4 same
hardware revision gate ro.revision above 4, else dv file none

(V for sc2, A for the nap column from sc-hardware-identity.md.)

Item Finding
Chip Broadcom BCM43526 on USB. Evidence: kernel module bcm_dbus.ko (“dbus” is Broadcom’s USB dongle bus layer) with 14 USB aliases, no PCI or SDIO aliases; the nvram header comment #BCM943526USB NVRAM for SKY85803 and A2 chip; firmware banner Broadcom BCM%d 802.11 Wireless Controller.
Firmware lib/firmware/brcm/bcm43526-firmware.bin, 335,872 bytes (0x52000), starts with the 4 bytes HDR0 (a Broadcom download header); embedded version string 6.37.114.64 (r516176); runtime banner format RTE (%s-%s%s%s) %s on BCM%s r%d ... (Broadcom’s dongle runtime).
NVRAM lib/firmware/brcm/bcm43526-ffff-ffff.nvm, 3,570 bytes, text key=value. Header: sromrev 11, boardtype 0x623, vendid 0x14e4, devid 0x43a0, boardrev 0x1452, aa2g=0x3 and aa5g=0x3, txchain=0x3, rxchain=0x3 (two chains each way), ccode=0, per-chain pa2ga*/pa5ga* power amplifier tables and maxp5ga*, noise level and rx gain tables, LED behavior bytes and wpsgpio=11, boardnum=${serno} as a template slot. The file also has a macaddr= line; I did not read its value. The 5 GHz tables are populated, so the nvram describes a dual-band RF section even though the software caps bands (below).
File naming bcm_dbus.ko builds the names brcm/bcm<chip>-firmware.bin and %s-%2x-%2x.nvm (the ffff-ffff in the shipped file is the fallback with no board ids). request_firmware appears in bcm_dbus.ko and not in wl.ko (control: strings finds both in bcm_dbus.ko). So the USB bus module requests the files; bcmdl is not called by any script in the volume (/usr/bin/grep -rl -a bcmdl . returns only the binary itself). (V, I)
Driver lib/modules/extra/<variant>/wl.ko (license Proprietary, depends bcm_dbus, vermagic 3.4.11+ SMP preempt mod_unload modversions ARMv7) banner Broadcom BCM%04x 802.11 Wireless Controller 6.37.114.64 (r516176); same version as the firmware. Build stamps in the module: Oct 24 2017 at 09:21:47 (plain) and 09:22:14 (-extra), plus a source stamp 2014-11-19 11:24:40 in both.
bcm_dbus.ko Identical in both variant directories (md5 matches, 79,816 bytes). License GPL. USB aliases cover Broadcom 0a5c (bd12, bd13, bd16, bd17, bd1d, bd1e, bd1f, bd23, 0bdc, 0000) plus 043e:3001, 1eda:0bdc, 050d:d321 and 0846:9011.

lib/modules/extra/bcm43526/wl.ko is 1,374,492 bytes and lib/modules/extra/bcm43526-extra/wl.ko is 1,388,324 bytes (13,832 bytes larger). A strings diff shows the -extra build adds: chanim_state, chanim_mode, chanim_ccathres, chanim_glitchthres, chanim_bgnoisethres, chanim_sample_period, chanim_threshold_time, chanim_max_acs, chanim_lockout_period, chanim_acs_record, chanim_stats (channel interference measurement and automatic channel selection), PKTDELAY_IND, txdelay_params, phy_rxiqest. Everything else in the diff is symbol suffix renumbering and the build time. Same firmware banner. So -extra is the same Broadcom driver version with the channel-quality and packet-delay instrumentation compiled in. etc/boxinit.d/50-wifi.rc sets BCM_MODULES_DIR_NAME bcm43526-extra for wifid, so the -extra variant is the one that runs in normal operation; the plain variant is what usr/bin/radio_test.sh loads (insmod ./lib/modules/extra/bcm43526/wl.ko) for RF test mode. (V)

  • usr/sbin/bcmdl (35,572 bytes): Broadcom’s USB dongle download tool, usage text “downloads a firmware image into an USB dongle”, options -n <nvram>, -i <pid>, strings “brcm RDL”. Not used by a script here (above).
  • usr/sbin/bcmwl (580,568 bytes): Broadcom’s wl command line utility renamed. usr/bin/radio_test.sh drives it for continuous-TX RF tests (pkteng, country, 2g_rate, chanspec); it hard-codes band b and a 2.4 GHz channel 6 default.
  • usr/lib/wifid/bcmdriver.so: the wifid backend for this chip. Loads bcm_dbus.ko then wl.ko from /lib/modules/extra/%s/, toggles user_gpio/RESET_WIFI. It also carries a 43602 branch (firmware_path=/lib/firmware/%s/firmware.bin nvram_path=/lib/firmware/%s/bcm943602mc_parrot.nvm, /lib/modules/%s/dhd.ko) for drones that use a BCM43602; that branch does not match any file in this volume. Constants 0x14e4 and 0xaa52 occur; I did not trace their use (O).
  • usr/lib/libnetmon-bcm43526b.so (network-monitor package drones/network-monitor-bcm43526b): wifid’s Broadcom control library: band and channel get/set (netmon_band_*, netmon_channel_get), RSSI and rate reads, netmon_bcm_iovar_get/set, event registration, and the YAML tuning loader, which reads /etc/wifi/%s_config.yaml (“cfg_custom: product %s”) with the product name taken from ro.hardware (the same string /etc/default.%s.prop uses). Only one default.*.prop exists (mppboard), so I infer the file used is mppboard_config.yaml. (V strings, I selection)
  • usr/share/netmon/{netmon_config,device_perf,country_regul}.yaml: the generic tuning tables (country_regul.yaml is 686 lines of per-country channel rules, device_perf.yaml has per-device rx gain and tx power by channel starting with a “Bebop” entry). Not decoded further.
  • etc/libshsettings/wifi.cfg: outdoor = true; autocountry = false; country = "US"; autoselect = true; band = 4; channel = -1;. (V) Meaning of band = 4 not decoded (O).

etc/wifi/*.yaml (V; no secrets in any of them)

Section titled “etc/wifi/*.yaml (V; no secrets in any of them)”
File Content
mppboard_config.yaml (918 bytes) This product. 802.11: ampdu_rx_wsize: 2 (comment: 1 disables AMPDU and fixes command lag, “should not happen on mpp given the high number of packets from the drone”, 5 would restore “the same behaviour than SkyController”), rts_enable: 1 (RTS/CTS on every frame), slot_time_us left commented (25 µs). bcmver: 44: US with the comment “this forbids 5GHz in US”. The comment calls the original SkyController a separate product from this one.
evinrudeboard_config.yaml (90 bytes) bcmver: 1: JP. A product named evinrude; no 802.11 block.
mykonos4_config.yaml, paros_config.yaml (645 bytes each, byte-identical in content) rx_chains 7, tx_chains 7 (bitmask, three chains), slot_time_us: 20, AMPDU rx/tx and window set to -1, with comments about BCM43602 (“unlike 43526, driver cannot be fixed with window size”). These are for drones that use a BCM43602, not this unit.

The comments confirm two Broadcom parts in Parrot’s tree (43526 and 43602) and that this unit is the 43526 case. (V)

Method: each negative was run twice, first against the sc2 tree and the same command against the nap tree as the positive control, using find -iname for names, /usr/bin/grep -rIl -a -i for contents, and modules.builtin for kernel features (control: 11 drivers/parrot entries found in sc2’s own modules.builtin). nap: counts are files found. (V)

Component sc2 nap (control) Reading
HUD daemon hudd, libhud, Qt5 libs, fonts, EGL/GLES, Mali, UMP, HX270 decoder 0, 0, none, none, none 5, 4, present No display UI in this build. (V)
Framebuffer and video: fb.ko, avifb, avicore, siihdmi.ko, cfb* in modules.builtin none all present No framebuffer, no HDMI transmitter driver. (V)
HDMI: hdmi_hotplug, 99-hdmi.rules none present The only “hdmi” in the volume is the string hdmi_plugged in usr/bin/mppd (shared code, no HUD to start) and usb.ids. No HDMI. (V)
GPS: gpsd, libgps, btgpsdemux, gps_config.txt, gps_firmware.bin, ti-st gps_drv none present usr/lib/libmppd_hw_mpp.so function mppd_hw_is_gps_supported at Thumb 0x4bc4 is movs r0,#0 ; bx lr (returns 0). The nap library libmppd_hw_nap.so has the same function at 0x3fbc returning 1. mppd contains “GPS module not supported by this hardware” for that case. (V)
Bluetooth: any BT kernel module, bluez, hciattach, rfkill, ttyPA1 UART mapping none ttyPA1 and btgpsdemux present in nap No Bluetooth stack in the volume. Not hardware proof by themselves: boxinit.rc still reads /factory/bluetooth_mac_address.txt into ro.factory.bt.mac_address, and pal-drivers.so contains a bluez HCI socket UART path (linux_uart_bt.c). Both are shared template code and do not show hardware. A BT MAC provisioned at the factory is possible (O). (V absence, O for chip)
Wi-Fi access point: hostapd, wpa_supplicant, start_dnsmasq_wlan1, 60-wl18xx.rules none (the dnsmasq binary exists but no service) present mppd_hw_has_wifi_ap at 0x4bc8 returns 0 here and 1 in nap. This unit does not run an AP: 50-wifi.rc starts only wifid --mode STA --ip 192.168.42.3 --dhclient. The phone link is USB (iAP/AOA), not Wi-Fi. (V)
Second radio (Atheros AR6004 or TI WL18xx) none present One radio only. (V)
5 GHz mppd_hw_get_wifi_bands_supported at 0x4bd0 returns 1; nap returns 3 I read this as a bitmask with bit 0 = 2.4 GHz and bit 1 = 5 GHz, so 1 = 2.4 only here and 3 = both on nap. The bit assignment is inference; the mppboard_config.yaml comment that bcmver 44: US “forbids 5GHz in US” and the nvram’s populated 5 GHz tables are the two hints that the radio itself can do 5 GHz while the product does not use it. (V values, I meaning)
Display panel, touch, backlight none none either Nothing in either image names a panel or backlight driver (the backlight string hits are in busybox and the manifest in both). No display here by the above. (V)
Audio none: no snd modules in modules.builtin, no codec in builtin or module lists None found. (V)

Present, so not absent: IMU and magnetometer. etc/boxinit.rc sets ro.sensors.imu.name "icm20608" (pname irq/4-icm20608_, trigger icm20608-dev0) and ro.sensors.magneto.name "ak8963" (pname irq/5-ak8963_co). The nap image has mpu6050 (irq/6, irq/7) in the same lines. modules.builtin carries drivers/iio/imu/inv_mpu6050/inv-mpu6050(-i2c).ko and drivers/iio/magnetometer/ak8975.ko as built-in, with industrialio-triggered-buffer, kfifo_buf and iio-trig-loop (so buffered sampling works through libiio, which sensorsd-mpp links). The legacy mpu6050*.ko and akm8975.ko modules are in the loadable set but no script loads them. sensorsd-mpp runs as a service (60-sensorsd-mpp.rc) and does magnetometer calibration (persist.magneto.bias). Which exact InvenSense part the in-kernel inv-mpu6050 driver treats as icm20608 cannot be seen without the kernel (O). (V)

Vermagic on every module here: 3.4.11+ SMP preempt mod_unload modversions ARMv7 (V). modules.devname is empty. modules.builtin has 140 entries.

Loadable modules in lib/modules/3.4.11+/kernel (28)

Section titled “Loadable modules in lib/modules/3.4.11+/kernel (28)”
Module Role (from its own description= or alias)
arch/arm/oprofile/oprofile.ko OProfile system profiler (debug)
crypto/ansi_cprng.ko, crypto/michael_mic.ko software PRNG; Michael MIC (TKIP)
drivers/input/misc/akm8975.ko AKM8975 compass driver (legacy; the board uses ak8963 through built-in IIO)
drivers/input/misc/mpu6050.ko, mpu6050_accel_gyro.ko, mpu6050_i2c.ko legacy TI-authored MPU6050 input drivers (dependency chain i2c, base, accel_gyro); not loaded by any script
drivers/input/touchscreen/usbtouchscreen.ko generic USB touchscreen driver; no hardware implied
drivers/mtd/devices/m25p80.ko SPI NOR flash
drivers/mtd/mtd_blkdevs.ko, mtdblock.ko, mtdblock_ro.ko, mtdchar.ko MTD block and char access
drivers/net/phy/libphy.ko, broadcom_br.ko PHY library; Broadcom BR PHY (alias for one MDIO id)
drivers/net/usb/rndis_host.ko USB RNDIS host (phone tethering style links)
drivers/net/wireless/hostap/hostap.ko Host AP common routines
drivers/parrot/i2c/muxes/p7-i2cmux.ko Parrot P7 pin-muxed I2C buses
drivers/scsi/scsi_wait_scan.ko SCSI scan wait
drivers/usb/gadget/g_aea.ko Apple External Accessory iAP2 USB gadget
drivers/usb/serial/pl2303.ko Prolific PL2303 USB serial adapter
net/ipv4/netfilter/nf_nat_ftp.ko, net/netfilter/nf_conntrack_ftp.ko FTP NAT and conntrack helpers (the unit serves FTP)
net/wireless/lib80211.ko, lib80211_crypt_{ccmp,tkip,wep}.ko common 802.11 crypto helpers

Of these, only g_aea.ko is loaded by a script (etc/boxinit.d/98-iosd.rc). The rest are available for manual use or are inherited from the Parrot kernel config. (V)

Section titled “g_aea.ko and the iOS accessory link (V, with inference marked)”

Description “Apple External Accessory iAP2”, author David Brownell (the Linux gadget serial lineage), parameters app_protocol, idVendor, idProduct, bcdDevice, iManufacturer, iProduct, iSerialNumber. Strings: ttyGS%d, iAP Interface, gser_aea, Parrot, Apple External Accessory. etc/boxinit.d/98-iosd.rc loads it with app_protocol=com.parrot.dronesdk idProduct=0x5040. It is a CDC-serial-style gadget that exposes the iAP2 interface and creates a ttyGS port; lib/udev/rules.d/61-juba-iap-tty.rules watches ttyGS1 and publishes a tty_iap device to the JUBA device layer, and 60-juba-iap.rules handles an Apple device (VID 05ac, PID 12xx) on the first port of ci_hdrc.0 as an iPod. The userland is usr/bin/iosd plus usr/lib/acid.so, which carries acid_coproc_setupI2c, acid_auth_getCertificate, acid_i2c_* and acid_ipodout_*: an iAP stack that reads an authentication certificate from an I2C coprocessor (ro.iosd.coproc.i2c_dev is a property name read by iosd; its value is not set in any file I read). I take that to be the Apple authentication chip that every iPod accessory needs (I; R for the chip being Apple’s MFi part, confirmed by a part number on the board or the coprocessor id acid logs). iosd also reads ci_hdrc.0/role, vbus_always_on and udc/state to roleswitch the port, and the app protocol strings include com.parrot.freeflight3. (V strings, I)

Android phones use the other path: usr/bin/aoad (libusb) probes for Android Open Accessory and runs an accessory bridge. So the phone connector speaks iAP2 to iOS and AOA to Android, over USB. (V)

Built-in drivers (modules.builtin, 140 entries; V)

Section titled “Built-in drivers (modules.builtin, 140 entries; V)”

The kernel drivers compiled in, grouped by what they tell us:

  • Parrot P7 SoC: p7mu-gpio, p7-i2cm-core, p7muadc, p7temp (the p7-temperature device that libmpphw.so reads), acs3-sdhci (SD/MMC host), p7_pwm, p7mu-pwm, rtc-p7mu, p7-spi-common, p7-spim, p7-spis, and drivers/mtd/nand/cast-nand.ko (the NAND controller for the UBI volumes). The p7mu family is Parrot’s power-management unit. (V names, I that these identify the SoC family)
  • USB: ci_hdrc (ChipIdea device/host core, two controllers ci_hdrc.0 and ci_hdrc.1 appear in scripts), ehci-hcd, udc-core, usbserial, usb-storage, asix, cdc_ether, cdc_ncm, usbnet, net1080, cdc_subset, usbhid, uhid, hid, hid-multitouch.
  • Input and sensors: evdev, joydev, gameport, uinput, industrialio, industrialio-triggered-buffer, kfifo_buf, iio-trig-loop, inv-mpu6050, inv-mpu6050-i2c, ak8975, led-class, switch_class, gpio-pca953x (an I2C GPIO expander driver is built in; whether one is fitted is O), spidev, i2c-dev, i2c-mux, regmap-i2c, gpio and fixed regulators, mpcore_wdt (ARM MPCore watchdog), pl330 (DMA).
  • Storage: ubi, ubifs, squashfs, cramfs, fat/vfat/msdos/ntfs, loop, dm-mod, dm-crypt, mmc_core, mmc_block, sd_mod, ramoops, deadline-iosched.
  • Network: IPv4/IPv6, netfilter NAT and masquerade, tun, dummy, af_packet, diag modules.

Not present in this list (with the positive control that the same list does contain drivers/parrot and ubifs): no ti-st, fb, avi*, siihdmi, p7gpu, wl12xx, cfg80211, mac80211, mali, ump, hx270, ccm, ctr, seqiv. That diff against the nap list is what removes the TI radio, HDMI, GPU and decoder from this unit. Note that cfg80211 and mac80211 are absent as well: the BCM43526 uses the proprietary wl.ko (a WEXT-style driver with wl_iw_* functions), not the cfg80211 stack. (V)

bcm43526/{bcm_dbus,wl}.ko and bcm43526-extra/{bcm_dbus,wl}.ko, covered in the Wi-Fi section. Both vermagic strings match the main kernel.

  • etc/boxinit.rc (on early-boot): ro.product.model "Skycontroller 2", ro.product.model.id "0x090f", ro.updater.filename "mpp_update.plf", sensor names as above. Factory properties are loaded from files that are not in this volume (/factory/serial.txt, pcb_serial.txt, wifi_mac_address.txt, bluetooth_mac_address.txt, joystick_zero_value.txt, joystick_min_max_value.txt); I list the file names only.
  • etc/build.prop and etc/default.mppboard.prop (the same properties): ro.build.alchemy.product=mpp, ro.parrot.build.group=drones, project/product mpp, variant linux, uid=mpp-linux-1.0.7, version 1.0.7, original_uid=mpp-linux-1.0.7-rc4, builder host id ad8f8115cd87 (a container id, not the kernel’s), ro.boot.console=ttyPA3, ro.debuggable=1, service.adb.tcp.port=9050. The release is a debug flavour with a root console (askfirst /bin/sh) and adbd started when ro.debuggable=1.
  • The mppboard name: boxinit builds /etc/default.%s.prop from ro.hardware, and default.mppboard.prop is the only such file. I infer ro.hardware is mppboard, which is the value the kernel’s machine description (via /proc/cpuinfo “Hardware” or ro.boot.hardware) supplies. libpinst.so’s machine table has a separate entry Mpp board. (V strings, I)
  • The ARSDK id: the earlier nap report decodes 0x090f as SKYCTRL_2 in libarsdk.so; here the image’s own boxinit says 0x090f is “Skycontroller 2”. The nap image (0x0913, SKYCTRL_NG, “Skycontroller”, nap_update.plf) is therefore a different product, which settles open question 1 and 2 of the nap report for this unit. (V)

No pcbrev. ro.revision comes from the kernel command line (above) and gates three things: bottom board firmware choice (revision above 4), iosd disabled on revision 0, and nothing else I found (usr/bin/mppd reads ro.revision too; I did not trace what for, O). usr/bin/iosd, usr/bin/mppd, usr/lib/libmppd_hw_mpp.so and usr/lib/libmppbb.so also reference ro.bb.bootloader.version. (V)

Mount and partition facts in the image (V, with the relay)

Section titled “Mount and partition facts in the image (V, with the relay)”

etc/boxinit.rc mounts ubi0:factory on /factory (ro), ubi2:data on /data and ubi2:update on /update. With the relayed root on ubi1:system this makes three UBI devices: 0 factory, 1 system, 2 data plus update, matching the relayed mtdparts (Pfactory, Psystem, Pupdate) one to one. (V, A)

  • 52-bottomboard.rules: DFU 03eb:2fd9 runs bottom_board_update; 03eb:2406 runs gamepad_plugged.
  • 55-juba-gamepad.rules: input names mpp2_gamepad and qpnp_pon (typed mpp2_power), and USB 03eb:2406 (mpp_gamepad) and 03eb:2407 (nap_gamepad) as JUBA devices. The mpp2 and qpnp_pon rules are for a Qualcomm-based sibling product, not this one.
  • 56-juba-bb.rules: hidraw 2406 as mpp_bb, 2407 as nap_bb.
  • 60-juba-iap.rules, 61-juba-iap-tty.rules: Apple device and ttyGS1 handling for iAP.
  • 70-memtester.rules: when a USB device 19cf:ffff (Parrot’s VID) is plugged, run mpp_bb_cli stayon 1 and start memtester.sh; on removal stop it. usr/bin/memtester.sh runs memtester 20M in a loop, lights the LED magenta (mpp_bb_cli on 6), and blinks via memtester_led.sh on a failure. This is a factory RAM test fixture hook. (V)
  • 70-juba-storage.rules, 99-juba-usb-generic.rules, 10-juba-firmware.rules: storage, generic USB, and firmware loader hooks of the JUBA device layer.
  • No Wi-Fi interface naming rule (unlike nap’s 60-wl18xx.rules); the BCM interface keeps its default name. (V)

Libraries (V from strings and symbols; nothing guessed)

Section titled “Libraries (V from strings and symbols; nothing guessed)”
File What it is Evidence
usr/lib/libpuf.so (8,780 bytes) Update package reader, not a physically unclonable function Exports puf_new, puf_destroy, puf_get_version, puf_get_app_id, puf_get_target_id, puf_get_file_size, puf_extract_to_buf, puf_extract_to_file, puf_compare_version, puf_version_tostring, puf_version_fromstring. Imports libtar (tar_open, th_read, tar_extract_file) and zlib (gzdopen, gzread). Strings .plf, .tar, .tmp, a magic PLF! compare, and version formats %u.%u.%u, %u.%u.%u-alpha%u, -beta%u, -rc%u. Used by mppd, arsdk-cli, libmppd_hw_mpp.so, libarsdk.so. It inspects update files (Parrot .plf header and tar.gz bundles such as mpp2_update.tar). What “PUF” expands to is not in the volume: O. (R: probably “Parrot update file”; a header comment in the drones/libpuf source would confirm.)
usr/lib/acid.so (437,204 bytes) iPod/iAP accessory protocol stack 1,335 exported text symbols: acid_ipodout_*, acid_music_*, acid_location_mbox, acid_app_mbox, acid_tagging_*, acid_coproc_setupI2c, acid_auth_getCertificate, acid_i2c_*, acid_hw_registerFileDescriptor, get_acid_priority_ipod_authentication; imports libusb, juba_*, pal_hid_*, pal_gpio_*, sup_* (the supporting runtime). Only usr/bin/iosd links it. Manifest project libmult/Acid, revision 50ffa888..., the same revision as in the nap manifest.
usr/lib/pal-core.so, pal-devs.so, pal-drivers.so, pal-juba.so, pal-utils.so Parrot platform abstraction layer (manifest os/PAL) pal_module_*, pal_log_dyn_*, pal_lib_desc_* (core); pal_hid_* plus packages/pal/devs/usb/ipod.c (devs); gpio, i2c, spi, hid and UART drivers with hard-coded nodes /dev/hidraw0, /dev/i2c-0, /dev/spidev1.1, /dev/usb-i2c, UART over bluez, TCP, unix socket and btgpsdemux (drivers); juba_* getters over udev data and /proc/mounts (juba); pal_lang_* (utils). The UART/bluez/btgpsdemux paths are code present in the library and not hardware present in this unit (see Absent).
usr/lib/plib.so (9,132 bytes) tiny hardware info helper plib_hwinfo_get_prod_name, _prod_full_name, _prod_rev, _device_tree_model, plib_gpio_n_set, plib_gpio_n_read, plib_i2c_write_read; reads /proc/cpuinfo (“Hardware”, “Revision”) and /proc/device-tree/model.
usr/lib/libpinst.so and pinst_trigger, pinst_version Parrot installer library (boot mode and partition handling) pinst_file_boot_*, pinst_gpt_boot_*, pinst_ubi_boot_*, pinst_ambaflag_*, pinst_get_board_info, pinst_get_board_cpu; machine name table with 60 plus entries including Parrot7 development board, Mykonos3 board, Milos board, Mpp board, FC7100 Workbench, Chimera board, mykonos4, paros, paros360, paros3d, NVIDIA Tegra124 PM375. Same table as the nap image (only a few bytes differ). The table is generic, so no single entry identifies this unit.
usr/bin/report-generic (8,060 bytes) macgyver (Parrot crash and event reporter) CLI report_generic <subject> [--output path], reads /etc/macgyver.conf, writes a SHA-1-tagged report under /data/generic. Used by 50-wifi.rc (broadcom-report after a Wi-Fi watchdog) and 98-usbctrl-mon.rc.
usr/bin/usbctrl-mon USB controller watchdog Monitors ci_hdrc.0/usb1/status, on a fatal state unbinds and rebinds ci_hdrc.0 and resets a USB PHY over /dev/i2c-0 (so a USB PHY or hub controller sits on I2C bus 0). 50-wifi.rc does the same unbind/bind for ci_hdrc.1 and pulses RESET_USB_HUB, which means a USB hub sits behind controller 1 together with the Wi-Fi chip (I).

etc/manifest.xml (107 projects, repo remote on a Parrot Gerrit host): pins os/kernel/linux to packages/linux-p7 at 76ecc73342... (the nap image pins 98244c00..., so a different kernel commit), os/p7/gpu and os/p7/vdec (listed but their output is not in this image), drones/bcm43526 at 58984037..., drones/network-monitor-bcm43526b, drones/wifid-bcmdriver, drones/mpp/atmel (packages/atmel-mpp), aftermarket/NAP/Atmel (atmel-nap, revision differs from nap’s), drones/mpp2/atmel, drones/mpp/products/{mpp,mpp2,nap}, the pinstrc trees build/pinstrc/{mpp,nap}, the Qualcomm trees (linux-msm8909, lk-msm8909, qcacld-2.0, qca6174-firmware, qcom-charger, mkbootimg), and drones/dragonfly/products/lk (a Little Kernel bootloader for the Parrot side). It is a shared tree for several products; a name there is not evidence of hardware here. (V)

Area sc2 (mpp, 0x090f, “Skycontroller 2”) nap (0x0913, “Skycontroller”)
Identity strings model “Skycontroller 2”, id 0x090f, mpp_update.plf model “Skycontroller”, id 0x0913, nap_update.plf
Build 1.0.7, Oct 24 2017, kernel commit 76ecc733... 2.0.0-rc3, Aug 9 2017, kernel commit 98244c00...
Bottom board AVR XMEGA, bottomboard.hex plus .dv.hex, 03eb:2406 “MPP GAMEPAD”, 22 buttons plus 5 axes, gated by ro.revision ATxmega32C4, xmega.hex 13,122 bytes, 03eb:2407 “NAP GAMEPAD”, 24 buttons plus 8 axes
Wi-Fi one Broadcom BCM43526 on USB, STA only TI WL18xx on SDIO as AP plus Atheros AR6004 on USB as STA
Display, HDMI, GPU, video decoder none HUD (Qt5 eglfs on Mali-400), Silicon Image HDMI transmitter, hardware decoder
GPS and Bluetooth none found GPS (ai2) and BT on one UART via TI shared transport
IMU icm20608 (irq 4 in thread name) mpu6050 (irq 6)
Magnetometer ak8963 (irq 5) ak8963 (irq 7)
Phone link iAP2 gadget plus I2C authentication coprocessor, AOA host, RNDIS AOA host, RNDIS; adb gadget over USB
Revision discovery ro.revision from the kernel command line (inferred) /sys/kernel/hsis/pcbrev default 00.00 (a table the nap kernel lacks)
Services iosd, usbnetd, usbctrl-mon, macgyver, memtester btgpsdemux, gpsd, dnsmasq, hudd
Hardware-hook answers (libmppd_hw_*): GPS, Wi-Fi AP, bands 0, 0, 1 1, 1, 3
Factory type mapping folders etc/mppd/mapping_{bebop,bebop_2,chimera,disco}.cfg (a different set of files than nap’s bebop/milos folders); the ro.factory.type selector exists in mppd (/etc/mppd/{ro.factory.type}) bebop or milos folders chosen by nap_factory_reader
Shared Linux 3.4.11+ board-file era kernel on ARMv7 (3.4.11+ SMP preempt), Parrot P7 driver set, boxinit/ulog/juba/pal userland, same JUBA udev conventions, shared atmel-mpp source revision same

(V from files in both trees; the nap column cites the earlier report where it was not re-read.)

Reading: the two images are different products on the same software base, and the nap image is the larger, richer device. That matches the nap report’s note that its own libarsdk keeps SKYCTRL_2 (0x090f, this product) separate from SKYCTRL_NG (0x0913). (V, I)

What a kernel image and the other partitions would still tell us, and what we cannot know from this volume:

  1. Pin map. The scripts use user_gpio names RESET_WIFI, RESET_USB_HUB, POWER_KEEP (plus whatever mppbb or others export); the numbers sit in the kernel’s board code (as they did in the nap kernel’s user-GPIO export function). The kernel image is the only place for them. The 107-project manifest names packages/linux-p7 only, with no board file path visible in this volume. (O)
  2. SoC and machine descriptor. The Parrot7 conclusion comes from driver names and temperature readers; the exact P7 revision is detected at runtime by the kernel and the machine name (mppboard, or something else) is in the machine descriptor and /proc/cpuinfo. libpinst.so lists Mpp board but also 60 other boards. The kernel image would show the descriptor and the MPP hw 0x...-style board-id log string if one exists. Confirming by bytes needs the kernel partition (the PLF section type 3 payload in an update file, or Pmain_boot dumped). (O)
  3. Board revision decode. ro.revision is a kernel command-line argument, so the kernel boot arguments in the bootloader or Pmain_boot give the revision strings. The mapping of revision numbers to board builds (EVT/DV/production) is not derivable here. (O)
  4. Why the bottom board has DV and main builds and what firmware version each is. Neither file has a version; the unit may report a version over the 16 byte vendor report, which the mpp_bb_cli and libmppbb.so paths would show if run, and the bootloader version is already available through dfu-programmer get bootloader-version. (O)
  5. The /factory partition. It holds the serial, PCB serial, Wi-Fi and BT MAC files, joystick zero and min/max calibration files and (likely) the ro.factory.type selector; none are in this copy by design. Whether a Bluetooth MAC file really exists on the unit, which would indicate a BT radio present but unused, can only be settled there. (O)
  6. /customer_prod, /data, /update: runtime state, country lock, persisted SSID and magnetometer bias. Not in the copy.
  7. Bootloader (Pbootloader 8M, Pmain_boot 16M): the manifest lists a Little Kernel product (drones/dragonfly/products/lk) for the Parrot trees. Its identity string, boot arguments and the mtdparts source are in those partitions. (O)
  8. The kernel configuration: configs.ko is in modules.builtin (kernel/configs.ko), so /proc/config.gz very likely exists on the live unit, which would give every driver option in one read. Do this with a live read only, which this task does not permit. (I)
  9. What the iAP coprocessor and the USB PHY or hub on I2C are by part number: only board photos or the I2C addresses in the kernel board table would tell. (O)
  10. Wi-Fi regulatory data and bcmver meaning beyond the yaml and nvram text. The Broadcom firmware is a closed binary and I read only its strings. (O)
  1. Frame layout of the 16 byte vendor report (LEDs, battery, RTC, calibration, firmware version). Needs either an AVR disassembly of the main hex or a live HID capture; the static route is to disassemble the HID handler in the flat image (hex address of the report ID 2 handler not located).
  2. What distinguishes the DV and main hex in function (feature set, not only layout): the main build carries more state variables and one more compare routine; which user-visible feature that is, is not decided.
  3. ro.hardware value and the machine name: confirm mppboard from the kernel image.
  4. Meaning of ro.iosd.coproc.i2c_dev and its value (not set in any file); and how iosd picks the I2C bus (the PAL driver hard-codes /dev/i2c-0 and /dev/usb-i2c).
  5. Whether 5 GHz is really disabled in this product: mppd_hw_get_wifi_bands_supported returns 1, the yaml says bcmver 44: US forbids 5 GHz in the US, and the nvram has 5 GHz data; the bit meaning and whether a country other than the US enables it needs the code in libnetmon-bcm43526b.so (netmon_bands_to_bcm) read in full.
  6. Constants 0x14e4 and 0xaa52 in bcmdriver.so and the bcmver index semantics (44 for US, 1 for JP).
  7. Whether any Bluetooth hardware exists in the unit (the volume has the MAC file name and PAL code only, no stack).
  8. Which InvenSense part the in-kernel inv-mpu6050 driver binds as icm20608, and the mounting matrix for either sensor.
  9. Whether /proc/config.gz is exposed, and the exact build time of the kernel versus the relayed 2017-10-24 (the rootfs build date is 09:10:16 UTC; the Wi-Fi modules were built at 09:21:47 and 09:22:14 the same day).
  10. What “puf” stands for and what “DV” stands for. Both have only inference or recollection behind them.

Comments

Spotted a mistake, or tried this on your own Bebop? Leave a note. Comments are kept on this site's own server.