Skycontroller 2 (mpp-linux-1.0.7): software stack, mapping and protocols
Static read of the live Skycontroller 2’s software stack (mpp-linux-1.0.7) by one agent, 2026-10-04. Status: Agent-cited, with these claims re-checked by the commissioning session against the copy: the unit’s own identity block and mapping files (action 26 bound in none of the four files, takeoff/land bound in all four: control in the same files), a single
Ardrone3_Piloting_Emergencystring inusr/bin/mppd(control:Ardrone3_Piloting_TakeOffalso once), and the video relay ports: the disassembly ofusr/bin/mppdbuilds asockaddr_inwithmovw r3, #0x1f1e(htons of 7711) stored beside the AF_INET word, and a second one withmovw r3, #0x8c13(htons of 5004), in this build at 0x24d74 and in thenapbuild at 0x249a0. Not re-derived: the emergency call site and its drone-manager state gate, the takeoff battery thresholds (7200 mV here against 10965 mV in the nap image), the key-to-button mapping, ARSDK counts and tables, the router rule table, and the binary inventory.Correction recorded: the earlier
napsoftware report andskycontroller-live-vs-image.mdsaid the port 7711 is not in the image. That was a method error: ARM code builds a port from a byte-swapped 16-bit immediate, which a search for the decimal number or for plain bytes cannot find. It is in both builds.Do not turn the emergency-action and takeoff-gate readings into public claims without a second read of
mppd.
Static analysis only. Image root R = the extracted system volume (bin sbin lib usr etc only). Every path below is relative to R unless it starts with nap:, which means firmware/extracted/skycontroller-nap/rootfs in the bebop-2 project. Nothing was connected to, no mcbebop tool was used, nothing was written under R. Scratch files (disassembly, string dumps, small decoders) are in the session scratchpad. No serial, MAC, Wi-Fi key or device id appears in this report.
Tags: Verified (I read the bytes, listing or disassembly), Agent-cited (taken from the relayed live readings or an existing note, not re-read), Inference, Open, Recollection (outside knowledge, with what would confirm it).
Address conventions. usr/bin/mppd is ET_EXEC, vaddr = file offset + 0x10000 for text and rodata (data segment +0x20000), disassembled as Thumb-2 (the file is stripped, so llvm-objdump --triple=thumbv7-linux-gnueabihf is needed; the default ARM-mode listing is garbage). Shared objects under usr/lib have vaddr = file offset for text; libarsdk.so data segment is +0x10000.
Negative results. Each one names the control I ran with the same tool and flags.
1. Summary
Section titled “1. Summary”- The image identifies itself as ARSDK product 0x090f:
etc/boxinit.rcsetsro.product.model "Skycontroller 2",ro.product.model.id "0x090f",ro.updater.filename "mpp_update.plf";etc/build.propsays projectmpp, uidmpp-linux-1.0.7, built 2017-10-24;libarsdk.soarsdk_device_type_str(0x2010c) decodes 0x090f toSKYCTRL_2. This is themppproduct, not thenapimage studied before. Verified. - The hub is still
mppd, but the hardware layer islibmppd_hw_mpp+libmppbb+libmpphw, the bottom board is themppboard (USB 03eb:2406, strings “www.parrot.com” and “MPP GAMEPAD” inlib/firmware/bottomboard.hex), and there is nohudd, no HDMI path, no GPS, no AP radio, no GPU/Qt stack. Sticks come from evdev (5 axes, -100..100), battery and LED and calibration go through hidraw report 2 (17-byte writes). Verified. - Mapping: four files
etc/mppd/mapping_{bebop,bebop_2,chimera,disco}.cfgplusdrone_manager.cfg(known_drones = ( );), flat inetc/mppd/, nobebop/milosfolders, no anafi4k file. The emergency action (26) is bound in none of them andBUTTON_Zdoes not exist in this button list; positive controls on the nap files (action 26 and BUTTON_Z both found with the same grep) confirm the negative. Bound actions: app_0, return_home, takeoff_land, video_record, take_picture, app_4, app_3, exposition inc/dec, and flips (not in disco). Verified for the default files; the unit’s own/factoryor/var/lib/mppdcopies are not in the image (Open). mppdstill sendsArdrone3_Piloting_Emergencyfrom exactly one place (0x20dae, case 26 of the action jump table), with no flying-state or battery check; the only gate in front of every button action is “drone manager state == 6 (DRONE_CONNECTED)”. Takeoff is refused (bothTakeOffandUserTakeOffpaths, 0x20d00 and 0x20d6a) when the controller battery state is 4 or 5, which libmppbb derives from voltage at or below 7200 mV (the nap threshold was 10965 mV). Verified.- Relayed reading checks:
GET /videoon TCP 7711 is Verified in the image (listen on 0.0.0.0:7711 at 0x24d74; the earlier nap report missed it because sockaddr ports are stored byte-swapped, see section 4.1); the SDP body isc=IN IP4 addr:192.168.53.1/m=video 55004 RTP/AVP 96/a=rtpmap:96 H264/90000; RTP leaves from UDP 5004 to the requester’s IP port 55004. mDNS service type_arsdk-090f._udp, name fromro.product.model, port 44444, all Verified; the USB-NCM link is not explained by the image (Open). - The kernel key codes 288..294 map, in this image’s own table (
libmppd_hw_mpp.so0x3702 to 0x3814), to logical buttons A, B, C, X, Y, R1, L1; with the default files that is app_0, return_home, takeoff_land, video_record, take_picture, app_3, app_4. That fits the relayed names for 288 (settings), 289 (return home), 290 (takeoff/landing), 293 and 294 (triggers); the relayed “B” and “A” for 291 and 292 are the video and photo actions by position (Inference). Verified table, Inference for the physical labels. - ARSDK:
libarsdk.sois a newer revision (c5d7b85f) than the nap one: 609 command descriptors (nap 603), same 17 projects,Skyctrl80 commands (nap 77, adds StartCalibration, AbortCalibration, MagnetoCalibrationStateV2), media and update tables gain Disco and Swing entries and the 0x0915 update file ismpp2_update.tar.gz. 0x0913 is still absent from both product tables. Verified. - Corrections and open items: the nap report’s Open item “port for GET /video” and its 7711 negative are resolved by this analysis (the same code is in the nap
mppdat 0x249a0, 0x249b6 and 0x24ad0); the published mapping page is accurate for the nap image but none of its tables apply to this image (section 8). Open: board button-to-bit assignment,/factoryoverrides, router rule table, HUD telemetry sink, USB-NCM path, whetherro.hardwareselectsdefault.mppboard.prop(which would start adbd and a console).
2. Architecture
Section titled “2. Architecture”2.1 Block diagram in prose
Section titled “2.1 Block diagram in prose”Boot is boxinit (sbin/boxinit, reads etc/boxinit.rc, etc/boxinit.dragonbase.rc and every etc/boxinit.d/*.rc). on boot starts class core (udevd, jubamountd, and adbd and a console only when ro.debuggable=1), on boot-main then starts class main and class mpp (etc/boxinit.rc:96-102). Class main: wifid, avabye, usbnetd, sensorsd-mpp, inetd, macgyverd, ulogcatd, kmsgd, crashdumpd, usbctrl-mon (the two memtester helpers are declared disabled). Class mpp: mppd, aoad, iosd. Verified (etc/boxinit.d/*.rc).
Toward the pilot’s hands: an Atmel ATxmega32C4 bottom board enumerates as one USB HID interface, VID 03eb PID 2406, with a 22-button + 5-axis input report (id 1) and a 16-byte vendor report (id 2). udev rule lib/udev/rules.d/55-juba-gamepad.rules tags the evdev node mpp_gamepad and 56-juba-bb.rules tags the hidraw node mpp_bb; libjuba hands both to clients by tag, not by path. The kernel evdev node is read by libmppd_hw_mpp (via libevdev); the hidraw node is opened O_RDWR by libmppbb for battery, LED, thresholds, calibration, RTC and power control. The only physical status indicator driven from the host is one RGB LED, and it is driven through the board (report 2 command bytes), not through sysfs GPIO as on the nap. Verified.
Toward its own sensors: mppd contains the sensor client in libmppd_hw_mpp (sensors_*, pomp client of abstract @sensorsd), served by usr/bin/sensorsd-mpp (libiio; IMU ICM-20608 and magnetometer AK8963, named in etc/boxinit.rc:37-41). Verified for names, Inference for the framing.
Toward the drone: wifid --mode STA --ip 192.168.42.3 --dhclient with WIFID_DRIVER=bcmdriver (Broadcom BCM43526, driver under lib/modules/extra/bcm43526*, tools usr/sbin/bcmwl, bcmdl, usr/lib/wifid/bcmdriver.so). mppd’s drone_manager module scans, connects (WIFID_IPC_* events over @wifid), then proxy_drone runs an ARSDK net backend toward 192.168.42.1 (immediate 44444 at 0x2a6d0, string “192.168.42.1” referenced at 0x2a6da). There is no hostapd/wpa_supplicant in this image and mppd_hw_has_wifi_ap returns 0, so the phone cannot join a controller-hosted AP. Verified.
Toward the phone, three paths exist in mppd and the image supports only the wired ones: (a) USB to Android through aoad (libaoa; accessory socket @/com/parrot/aoad/<bus>/<dev>; mppd runs libmux on it); (b) USB to iOS through iosd (loads g_aea.ko with app_protocol=com.parrot.dronesdk idProduct=0x5040, switches ci_hdrc.0 role host/gadget, sets property iosd.charge.device) and mppd opening /dev/ttyGS1; (c) an ARSDK net backend (proxy_mobile_net) listening on TCP 44444 and published by mDNS, reachable over any eth*/usb* interface that usbnetd and avabye manage. Verified for each program and string; the NCM link seen live is Open (2.4).
Video: ARStream2 from the drone is bridged as raw UDP (no libvideo-streaming, librtp or libh264 is linked into mppd; readelf -d NEEDED list in section 6). Section 4.1 has the ports. A second restream copy still goes to 127.0.0.1:54321 for a HUD that does not exist in this image. Verified.
2.2 Link table
Section titled “2.2 Link table”| From | To | Transport | Framing | Evidence | Tag |
|---|---|---|---|---|---|
| bottom board | kernel HID | USB FS, 03eb:2406, EP 0x81 IN and 0x02 OUT interrupt, 32 B, bInterval 10 | report 1 (22 btn + 5 axes), report 2 (16 B vendor) | lib/firmware/bottomboard.hex config descriptor, HID report descriptor |
Verified |
| evdev (name relayed as “www.parrot.com MPP GAMEPAD”) | mppd via libmppd_hw_mpp |
/dev/input/event*, juba tag mpp_gamepad |
EV_KEY, EV_ABS | 55-juba-gamepad.rules, libmppd_hw_mpp.so 0x36f4 (libevdev_next_event) |
Verified |
| hidraw | mppd via libmppbb |
/dev/hidraw*, juba tag mpp_bb, O_RDWR |
report id 2, 17-byte writes | 56-juba-bb.rules, libmppbb.so 0x1645, 0x14f8 |
Verified |
sensorsd-mpp |
mppd (libmppd_hw_mpp) |
abstract @sensorsd, libpomp |
SYNC, attitude, MAGNETO_CALIB messages | strings in libmppd_hw_mpp.so |
Verified strings |
wifid |
mppd |
abstract @wifid (and @wifid-%s), libpomp |
typed messages | strings in mppd |
Verified strings |
mppd |
drone | Wi-Fi STA wlan, TCP 44444, 192.168.42.1 | ARSDK net backend | movw 0xad9c at 0x2a6d0 | Verified |
mppd |
phone (net) | TCP 44444 listen, mDNS _arsdk-090f._udp |
ARSDK net backend | movw 0xad9c at 0x27fb0 (listen) and 0x27622 (publisher) | Verified |
mppd |
phone (video request) | TCP 0.0.0.0:7711 | HTTP/1.x GET /video |
0x24d74 (movw 0x1f1e, stored swapped) then pomp_ctx_listen at 0x24d9e |
Verified |
mppd |
phone (video data) | UDP src 5004, dst peer:55004 | raw ARStream2/RTP payload relay | bind at 0x24dea (0x8c13 swapped), send at 0x237ee (0xdcd6 swapped) | Verified |
mppd |
HUD (none) | UDP 127.0.0.1:54321 | restream copy | 0x24ea4 (0x31d4 swapped), string “127.0.0.1” at 0x2fbde | Verified numbers; no consumer binary in image |
mppd |
phone (USB Android) | @/com/parrot/aoad/%d/%d |
libmux | usr/bin/aoad, mppd strings |
Verified strings |
mppd |
phone (iOS) | /dev/ttyGS1, raw |
libmux | strings “cannot open /dev/ttyGS1”, “iap tty opened” | Verified strings |
mppd |
updater | USB stick, ro.updater.watch.dir, mux updater channel |
puf container, md5 |
strings | Verified strings |
2.3 What replaced what
Section titled “2.3 What replaced what”| nap component | This image | Tag |
|---|---|---|
hudd, libhud*, Qt5, Mali, hx270, hdmi_hotplug, libvideo-* |
none of them are in R; mppd keeps hud_update_* senders and the 127.0.0.1:54321 copy |
Verified |
libmppd_hw_nap |
libmppd_hw_mpp (gamepad, LED module, shutdown, sensors, capability constants) |
Verified |
libnapbb (HID battery plus GPIO bar LEDs 55 to 58) |
libmppbb (HID only: battery, RGB LED, thresholds, stay-on, auto-reboot, RTC, calibration, raw mode) |
Verified |
nap_bb_cli |
mpp_bb_cli (usage in section 6.3) |
Verified |
nap_factory_reader (sets ro.factory.type, default SSID) |
none; ro.factory.* come from etc/boxinit.rc ${file:/factory/...} lines; nothing sets ro.factory.type |
Verified |
gpsd, btgpsdemux, libgps |
none; mppd_hw_is_gps_supported returns 0 |
Verified |
hostapd, wpa_supplicant |
none; mppd_hw_has_wifi_ap returns 0; wifid is Broadcom-driver based |
Verified |
libvideo-streaming/librtp in mppd |
none linked; relay is plain UDP | Verified |
bottom_board_update flashing xmega.hex |
flashes bottomboard.hex if ro.revision > 4, else bottomboard.dv.hex; gamepad_plugged runs when 03eb:2406 appears |
Verified |
plf update container |
puf (libpuf.so, strings “bad puf target”, “puf md5sum mismatch”) |
Verified |
| none | usbnetd (DHCP on eth*,usb*, dnsmasq server on 192.168.N.1 or udhcpc client), iosd, macgyverd, mppinput, usbctrl-mon |
Verified strings |
2.4 Two unresolved links
Section titled “2.4 Two unresolved links”- USB-NCM.
lib/modules/3.4.11+/modules.builtinlistscdc_ncm.ko,cdc_ether.ko,usbnet.koas built in, andusbnetddefaults toINTERFACES=eth*,usb*, so a USB network interface would be managed automatically (DHCP server on 192.168.N.1, orudhcpc). The only gadget module shipped isg_aea.ko; nog_ncm/g_etherfile exists inlib/modules. So the image shows the host side machinery but not how this unit would present NCM as a gadget. Open (negative control:modules.builtingrep forusbreturns 15 lines, the same grep forncmreturns the one cdc_ncm line). etc/default.mppboard.propcontainsro.debuggable=1,ro.boot.console=ttyPA3,service.adb.tcp.port=9050.sbin/boxinitcontains the strings/etc/default.prop,/etc/default.%s.prop,/etc/build.prop, so adefault.<ro.hardware>.propis read.etc/boxinit.rc:14-15starts a console andetc/boxinit.d/50-adb.rcstartsadbdwhenro.debuggable=1. Whetherro.hardwareismppboardon a running unit comes from the kernel command line, which is not in R. Open.
3. Mapping files and the emergency binding
Section titled “3. Mapping files and the emergency binding”3.1 Files and format
Section titled “3.1 Files and format”etc/mppd/ contains drone_manager.cfg (20 bytes, known_drones = ( );) and four mapping files of 4,236 to 4,238 bytes (disco 3,858). libconfig syntax, root group mapping, version = 1, arrays buttons, axes, expo, invert. Verified.
The loader (function around 0x19200 to 0x19640 in mppd) walks a three entry table in .data.rel.ro at 0x4401c: /factory, /etc/mppd/{ro.factory.type}, /etc/mppd. For each, it builds <dir>/<module>.cfg, opens it, and on ENOENT or a failed template expansion goes to the next entry (log “Property %s not found”); on the first file that opens it reads it; the strings “Copying default config %s to config %s” and /var/lib/mppd indicate that it is copied to /var/lib/mppd/<module>.cfg and that copy is what later boots read (Inference, as in the nap report). Because nothing in this image sets ro.factory.type (a recursive /usr/bin/grep -rla ro.factory.type over R matches only usr/bin/mppd; control: the same grep for ro.factory.serial matches etc/boxinit.rc, mppd, iosd, wifid), the template entry fails and the effective default dir is /factory if it holds the file, otherwise /etc/mppd. Verified for the order and the failure branch (disassembly 0x19350 to 0x19548); the contents of /factory and any earlier /var/lib/mppd copy are Open.
Selection by drone: mppd 0x1f736 maps the drone’s ARSDK product id minus 0x901 through a 22-byte table to a slot; the slot name array at 0x44080 is mapping_bebop, mapping_bebop_2, mapping_disco, mapping_chimera, value 4 means unsupported (“Bad product %04x, using default product instead”). Decoded: 0x901 BEBOP to bebop, 0x90c BEBOP_2 to bebop_2, 0x90e EVINRUDE to disco (Evinrude is the Disco only by Inference), 0x916 CHIMERA to chimera; every other id including 0x914 ANAFI4K and every Skycontroller id is unsupported. Verified. The nap table also had ANAFI4K as slot 3; it is gone here.
3.2 Action, button and axis lists in the files
Section titled “3.2 Action, button and axis lists in the files”Header comment in buttons: actions 0 to 15 app_N, 16 return_home, 17 takeoff_land, 18 video_record, 19 take_picture, 20 camera_exposition_inc, 21 camera_exposition_dec, 22 flip_left, 23 flip_right, 24 flip_front, 25 flip_back, 26 emergency, 27 center_camera. There is no 28 (cycle_hud) in the comment, although the loader accepts actions up to 29 (cmp r5, #0x1d; bls at 0x1dcdc) and the action handler still has the cycle_hud branch (0x20bb0, tests 0x1c) that writes hudd.visibility.level. Verified.
Button bits: 0x1 A, 0x2 B, 0x4 C, 0x8 X, 0x10 Y, 0x20 L1, 0x40 R1, 0x80 START (comment says “not mappable”), 0x100 LEFT_JS, 0x200 RIGHT_JS, 0x400 L2, 0x800 R2, 0x1000 VIRTUAL_LEFT_L, 0x2000 VIRTUAL_LEFT_R, 0x4000 VIRTUAL_LEFT_U, 0x8000 VIRTUAL_LEFT_D, 0x10000 to 0x80000 VIRTUAL_RIGHT_L/R/U/D, 0x100000 VIRTUAL_SLIDER_L, 0x200000 VIRTUAL_SLIDER_R. 22 bits. There is no BUTTON_Z. Verified (/usr/bin/grep -c BUTTON_Z is 0 in all five files, control: 2 lines in nap:etc/mppd/bebop/mapping_bebop.cfg). The same bit order is the order in libmppd_hw_mpp.so gamepad_button_name (tbb decode: index = bit position, A=0 through VIRTUAL_SLIDER_R=21), so the files and the library agree.
Axes: 0 AXIS_LEFT_LR, 1 AXIS_LEFT_UD, 2 AXIS_RIGHT_LR, 3 AXIS_RIGHT_UD, 4 AXIS_SLIDER_LR (5 axes; nap had 8 with MINI axes). Axis actions: 16 roll, 17 pitch, 18 yaw, 19 gaz, 20 camera_pan, 21 camera_tilt. Verified (and gamepad_axis_name decode agrees).
Special actions: none in the file format beyond the nap list; a button entry may combine bits (the flips use VIRTUAL_LEFT_x + BUTTON_X), and axes entries carry a buttons field that is 0 everywhere. START is not mappable per the comment.
3.3 What is bound (parsed, comments stripped)
Section titled “3.3 What is bound (parsed, comments stripped)”Parser result for the buttons arrays, positive control by the nap check below:
| File | Bindings (action to mask) |
|---|---|
mapping_bebop.cfg, mapping_bebop_2.cfg, mapping_chimera.cfg (identical bodies, 13 entries) |
0 app_0 to 0x1 (A); 16 return_home to 0x2 (B); 17 takeoff_land to 0x4 (C); 18 video_record to 0x8 (X); 19 take_picture to 0x10 (Y); 4 app_4 to 0x20 (L1); 3 app_3 to 0x40 (R1); 20 camera_exposition_inc to 0x800 (R2); 21 camera_exposition_dec to 0x400 (L2); 22 flip_left to 0x1008; 23 flip_right to 0x2008; 24 flip_front to 0x4008; 25 flip_back to 0x8008 (each is VIRTUAL_LEFT_{L,R,U,D} plus X) |
mapping_disco.cfg (9 entries) |
the first nine rows above, no flips |
axes: 18 yaw on axis 0, 19 gaz on axis 1, 16 roll on axis 2, 17 pitch on axis 3, 21 camera_tilt on axis 4 (the slider); camera_pan (20) is not bound. expo: axes 0 to 4 all 1 in bebop, bebop_2 and chimera, all 0 (linear) in disco. invert: all 0. The files differ from each other only in line 1 (product name) except disco (no flips, linear expo). Verified (diff of each against mapping_bebop.cfg).
3.4 Emergency (action 26)
Section titled “3.4 Emergency (action 26)”/usr/bin/grep -n "action = 26"over the four live mapping files returns nothing. Control, same tool and flags overnap:etc/mppd/bebop/*.cfg: one hit in each of the five files (line 97,buttons = 0x20; /* BUTTON_Z */). The wordemergencyappears in the live files only in the header comment (line 34). The parser output above also lists no 26. Verified.- There is no BUTTON_Z in the live button list (3.2) and no button that the relayed readings name as a safety or emergency key (keys 288 to 294, 7 keys).
- Other binding routes in code:
mppdaccepts runtime mapping changes from the phone (Mapper_Map_button_action, handleron_map_button_action_cmd_received, valid for actions 0 to 29), so any mapped button could be bound to 26 after the fact, and a changed mapping persists in/var/lib/mppd. Verified for the handler and range check, Inference for the persistence (same as the nap report’s reading). - Code side: the
Ardrone3_Piloting_Emergencydescriptor is loaded from exactly one site, 0x20dae. Method: resolve every GOT slot named inreadelf -rto its offset from the GOT base 0x4434c and search all literal-pool loads. Control: the same script on the napmppd(base 0x44358) reproduces the nap report’s site 0x20dca for Emergency and the other known sites (UserTakeOff three, Flip four).libmppd_hw_mpp.soandlibmppblackbox.sodo not name the Emergency descriptor (their descriptor strings are SkyControllerState_AttitudeChanged, CalibrationState_*, SettingsState_CPUID, and none for Blackbox). Verified. - Gating in front of the action: before the jump table (0x20c3a) the handler calls
0x1c7f8, which returns 1 only when the drone manager’s state field (offset 0x28) equals 6; the state enum order from the strings is IDLE, DRONE_SELECTING, WIFI_DISCOVERING, WIFI_CONNECTING, SDK_CONNECTING, DRONE_CONNECTING, DRONE_CONNECTED, so 6 is DRONE_CONNECTED (Inference for the enum order, Verified for the compare). Case 26 then encodes the command with no arguments and sends it. No flying-state, battery or confirmation check. A message from the phone can also reach the drone through the proxy (Inference, same router structure as nap). - Conclusion for the live unit’s defaults: no hardware button is bound to emergency in any of the four default files and none can be (no Z button). That holds for defaults only;
/factory/*.cfgis searched first and is not in the image (Open).
3.5 Comparison with the published page (nap image) and with the relayed readings
Section titled “3.5 Comparison with the published page (nap image) and with the relayed readings”What the published page says that differs here (the page is correct for the nap image):
| Page claim (nap image) | This image |
|---|---|
ten files in folders bebop/ and milos/, plus drone_manager.cfg |
four mapping files and drone_manager.cfg directly in etc/mppd/; no anafi4k file |
folder chosen by ro.factory.type (written by nap_factory_reader) |
nothing sets ro.factory.type; search order /factory, template (fails), /etc/mppd |
actions 0 to 28 including cycle_hud |
comment lists 0 to 27; handler still has 28 |
| 24 button bits: A,B,C,X,Y,Z,L1,R1 then VIRTUAL_LEFT/RIGHT/MINI | 22 bits: A,B,C,X,Y,L1,R1,START,LEFT_JS,RIGHT_JS,L2,R2 then VIRTUAL_LEFT/RIGHT/SLIDER; BUTTON_Z absent; L1 is 0x20 and R1 is 0x40 (nap 0x40 and 0x80) |
| 8 axes, MINI axes 4 to 7, camera_pan on axis 6, tilt on 7 | 5 axes, slider is axis 4 and carries camera_tilt; no camera_pan binding |
bebop/ binds 8 buttons including 26 emergency on BUTTON_Z, milos/ binds 7 and no emergency |
13 buttons (9 in disco), no emergency anywhere |
| expo 1 on axes 0 to 3, 0 on 4 to 7, same in all files | expo 1 on axes 0 to 4 in three files, 0 on all five in disco |
app_0 on L1 (bebop) or B (milos); return_home on Y; takeoff_land on X; video_record on C; take_picture on R1; center_camera on B or Z; cycle_hud on A or L1 |
app_0 on A; return_home on B; takeoff_land on C; video_record on X; take_picture on Y; app_4 on L1; app_3 on R1; exposition inc/dec on R2/L2; no center_camera, no cycle_hud, no app_N beyond 0, 3, 4 |
| flips not bound | flips bound to a virtual direction plus X (not in disco) |
| “which product number selects which file: Open” (resolved later in the nap report as the drone’s id) | same rule, table decoded above; ANAFI4K unsupported |
TAKE_PITCURE comment typo |
present in live comments too (mapping_bebop.cfg line 89); CYLE_HUD does not occur |
Relayed readings against the image:
| Relayed reading | Image | Tag |
|---|---|---|
| input device “www.parrot.com MPP GAMEPAD”, id 03eb:2406 | descriptor strings “www.parrot.com” and “MPP GAMEPAD”, device descriptor VID 03eb PID 2406 (bottomboard.hex 0x484c, 0x487c, 0x4888); that Linux joins manufacturer and product into the input name is Recollection of hid-core.c; the unit’s own evtest output would confirm it |
Verified bytes, Recollection for the join |
| axes -100..100 | report 1 axes X Y Z Rx Ry, logical min -100 max 100, 8 bits each (decoded at 0x47db to 0x47ef) | Verified |
| keys 288..294 | libmppd_hw_mpp EV_KEY table (0x3712 onward) |
Verified |
Which kernel keycode is which logical button, from the library (not guessed):
| evdev code | logical index (bit) | Name | Default action (bebop family) |
|---|---|---|---|
| 288 (0x120) | 0 | BUTTON_A | app_0 |
| 289 | 1 | BUTTON_B | return_home |
| 290 | 2 | BUTTON_C | takeoff_land |
| 291 | 3 | BUTTON_X | video_record |
| 292 | 4 | BUTTON_Y | take_picture |
| 293 | 6 | BUTTON_R1 | app_3 |
| 294 | 5 | BUTTON_L1 | app_4 |
| 295 | 7 | BUTTON_START | not mappable |
| 296, 297 | 8, 9 | LEFT_JS, RIGHT_JS | unbound |
| 298, 299 | 10, 11 | L2, R2 | exposition dec, inc |
| 300, 301, 302, 303 | 13, 12, 14, 15 | VIRTUAL_LEFT_R, L, U, D | with X: flip right, left, front, back |
| 720, 721, 722, 723 | 17, 16, 18, 19 | VIRTUAL_RIGHT_R, L, U, D | unbound |
| 724, 725 | 21, 20 | VIRTUAL_SLIDER_R, L | unbound |
| any other code | 22 | BUTTON_UNKNOWN | ignored |
Read from the library’s compare tree (cmp r3, #0x12a and similar at 0x3712 to 0x3806), a value of 1 is a press and 0 a release, 2 (repeat) is dropped. Verified. Axes: EV_ABS codes 0 to 4 map identically to axis indices 0 to 4 (a byte table at 0x5049 reads 0,1,2,3,4); anything above 4 is ignored. The value is rescaled as 100 - 200 * (v - min) / (max - min) using the device’s reported min and max (constants 200.0 and 100.0 at 0x39b8 and 0x39c0), so the logical axis is the negated evdev value when the kernel range is -100..100. Verified formula; the sign consequence is Inference from the relayed range. Optional circle shaping (SQUARE/CIRCLE modes, axis_squaring_circle, axis_bound_inscribed_circle) applies to the two stick pairs and apply_axis_expo_int to the slider. Verified names and calls.
Relay versus mapping, per key: 288 settings is app_0 (the app decides what it does), 289 return home and 290 takeoff/landing agree by name, 293 and 294 are R1 and L1 (app_3, app_4, the “triggers”), and 291, 292 are record and photo (the relayed names “B” and “A” are probably the button caps; Inference, and the physical key to HID bit assignment is in the board firmware, which was not decoded, Open).
Two caveats on reachability, both Inference with a Recollection: the board’s HID descriptor declares 22 button bits (usage 1 to 22). Linux hid-input.c (Recollection, would be confirmed by the unit’s evtest listing the supported codes) maps joystick button usages 1 to 16 to codes 288 to 303 and 17 and up to 704 and up. So codes 704 to 709 would reach the library as BUTTON_UNKNOWN, and the VIRTUAL_RIGHT and SLIDER codes (720 to 725) cannot be produced by this descriptor. The relayed list stops at 294, so the flip combinations (virtual left bits plus X) also have no source on this hardware as read so far.
4. mppd behaviour
Section titled “4. mppd behaviour”4.1 Listening and relay (video, ARSDK, mDNS)
Section titled “4.1 Listening and relay (video, ARSDK, mDNS)”Method note, and a correction to the nap report. A sockaddr is built with the port stored in network byte order, so port 7711 (0x1e1f) appears as movw r3, #0x1f1e, 5004 (0x138c) as 0x8c13, 54321 (0xd431) as 0x31d4 and 55004 (0xd6dc) as 0xdcd6. A search for the host-order immediates (44444 as 0xad9c, 5004, 55004, JSON ports) finds the numbers used in arsdk_* calls and JSON, but misses every raw sockaddr. The nap report used the host-order search, found 44444, 5004 and 55004 and concluded 7711 was absent; the same byte-swapped immediates exist in the nap mppd (0x249a0 movw 0x1f1e, 0x249b6 movw 0x8c13, 0x24ad0 movw 0x31d4), so the nap image listens on 7711 as well. Verified (the nap listing is nap-mppd.dis in the scratchpad).
start_restream (function at 0x24c24, listing 0x24d54 to 0x24f50):
| Socket | Address and port | Evidence |
|---|---|---|
restream request context (pomp_ctx_listen, TCP) |
0.0.0.0, port 7711 | memset then strh 0x1f1e at 0x24d74/0x24d7e, pomp_ctx_listen at 0x24d9e |
restream data context (pomp_ctx_bind, UDP) |
0.0.0.0, port 5004 | movw 0x8c13 at 0x24d8a, pomp_ctx_bind at 0x24dea |
| hud restream data context (UDP) | 127.0.0.1, port 54321 | inet_addr of the string at 0x2fbde, movw 0x31d4 at 0x24ea4, bind at 0x24eb6 |
Request handler restream_request_received (0x23900): it compares the first 17 bytes against GET /video HTTP/1 (strncmp at 0x23abe), sets system property mppd.restream.ip to the peer address from pomp_conn_get_peer_addr/inet_ntoa (log “restream requested: ip=‘%s’”), and answers with two asprintf templates.
SDP body, asprintf(&body, fmt, "192.168.53.1", 55004) at 0x23a22 (string at 0x2fc34 and 0x2fc41):
c=IN IP4 addr:192.168.53.1m=video 55004 RTP/AVP 96a=rtpmap:96 H264/90000Header template (0x2fcac, LF line endings only): HTTP/1.1 200 OK, Content-Type: application/x-www-form-urlencoded, Content-Length: %d, Connection: close, blank line, body. Verified. The addr: prefix inside the c= line is literal template text, not a syntax error in this report.
Property monitor restream_prop_event_cb (0x234c0) turns the property into a destination: empty means “disable restream” (0x23410), a non-empty value calls 0x23460, which strdups the IP into the context (offset 0x2c) if it changed (“enable restream: ip=‘%s’”). The stream data callback (0x237a0) then calls pomp_ctx_send_raw_buf_to on the data context (the one bound to UDP 5004) with destination inet_addr(ip) and port stored as 0xdcd6, which is 55004 (0x237ee). So RTP is sent from UDP 5004 to the requester’s IP, UDP 55004. With qos_mode equal to 1 the data socket gets setsockopt IP_TOS value 0x80 (0x2372a to 0x2373a; log “set ip/mac QOS mode”). Verified. Relayed reading “starts an RTP relay from port 5004”: Verified.
Not a pass-through for ARStream2 metadata: the drone-to-controller JSON announces arstream2_client_stream_port 55004, arstream2_client_control_port 55005, arstream2_supported_metadata_version 1 and the controller-to-phone JSON announces arstream2_server_stream_port 5004, arstream2_server_control_port 5005, arstream_fragment_size 65000, arstream_fragment_maximum_number 4, c2d_update_port 51, c2d_user_port 21 (both JSON strings are in mppd rodata). mppd forwards the payload as raw buffers (“stream bridge data created: %s:%d <-> %s:%d”, “send_stream_data”), and none of libvideo-streaming, libvideo-metadata, librtp, libh264 is linked. Inference: it bridges rather than parses. Verified for the link list.
ARSDK net backend: port 44444 (0xad9c), three sites: 0x27622 in arsdk_publisher_avahi_start (the published port), 0x27fb0 in arsdk_backend_net_start_listen (the mobile-facing listener, proxy_mobile_net), and 0x2a6d0 (the drone-facing connect with the string “192.168.42.1”). Verified. Which local address the listener binds to is in a sockaddr built at 0x27faa and was not decoded (Open).
mDNS: proxy_mobile_net calls arsdk_publisher_avahi_new/start with name ro.product.model (“Skycontroller 2”), id ro.factory.serial (fallback ro.serialno) and type strtol(ro.product.model.id) = 0x090f (code at 0x275a8 to 0x27630, property names resolved by literal-pool decode). libarsdk.so write_avahi_xml writes /etc/avahi/services/arsdk-ng.service (a symlink to /tmp/avahi/services, etc/avahi/services) from this template:
<service-group> <name replace-wildcards="yes">%s</name> <service> <type>_arsdk-%04x._udp</type> <port>%u</port> <txt-record>{"device_id":"%s"}</txt-record> </service></service-group>and usr/bin/avabye (config etc/avabye.conf: INTERFACE=eth*,usb*, SERVICE_DIR=/etc/avahi/services) answers for it. So the instance is Skycontroller 2._arsdk-090f._udp.local, SRV port 44444, TXT device_id = the serial. Verified for every link of the chain except avabye’s wire behaviour (not read, Agent-cited from the relayed mDNS reading).
Other listeners and ports found: mppd.mobile.debug property enables a TCP listener on 127.0.0.1 (port not decoded, Open); inetd runs ftpd -wS /var/lib/ftp on TCP 21 (etc/inetd.conf), which is the c2d_user_port 21 in the JSON (services analysis belongs to the network note). Verified for the text.
4.2 Takeoff gating, emergency, battery
Section titled “4.2 Takeoff gating, emergency, battery”The button action handler is 0x20b90 (log “Button Action %s(%d)” via string 0x2f2ad at 0x20baa). Action 28 is tested first (0x20bb0), actions 16 to 27 go through a 12-entry tbh at 0x20c64. Case sites resolved from GOT offsets: NavigateHome 0x20c9a, UserTakeOff 0x20d04 and 0x20d1a and 0x20d84, TakeOff 0x20d8c, Landing 0x20d90, Emergency 0x20dae, Flip 0x20db2/0x20dbc/0x20dc6/0x20dd0, VideoV2 0x20ddc, PictureV2 0x20df4. Flip direction arguments: left 3, right 2, front 0, back 1 (same as nap). Verified.
Takeoff gate: both takeoff paths load the controller battery state (context offset 0xb8), compute state - 4 and test <= 1 unsigned (0x20ce2 to 0x20cea and 0x20d4c to 0x20d54); if true they log “Refusing takeoff because MPP battery is critically low” (strings at 0x2f31a, referenced 0x20d00 and 0x20d6a) and do nothing. So takeoff is refused in states 4 (DISCHARGING_CRITICAL) and 5 (AUTO_SHUTDOWN). The nap mppd refused only state 2 (CRITICAL). Verified. State values and their voltages are in section 6.2. Landing and emergency are not gated.
Same as nap and not repeated: action 28 cycle_hud writes properties hudd.visibility.level (and reads hudd.visibility.level_count), which no process in this image consumes (no hudd). Inference for “no consumer” (no binary or string in R reads those properties; control: /usr/bin/grep -rla hudd.visibility matches only usr/bin/mppd, and the same grep finds mppd.restream.ip in mppd as expected).
4.3 What it forwards and consumes (controller side)
Section titled “4.3 What it forwards and consumes (controller side)”Ardrone3 and Common descriptors used by mppd (85, list in the scratchpad sc2-used.txt; nap 88): piloting (PCMD, TakeOff, UserTakeOff, Landing, NavigateHome, Emergency), flip, camera orientation and velocity (and states), media record and picture, exposition, media streaming enable, flying/attitude/position/alert/motion states, battery, wifi signal, country, current date and time. Controller-owned projects it speaks itself: Mapper (all 16), Drone_manager (7), Skyctrl AccessPointSettings and state, CoPiloting (SetPilotingSource, PilotingSource), CommonEventState Shutdown, CommonState AllStatesChanged, SettingsState (AllSettingsChanged, ProductSerialChanged, ProductVersionChanged), SkyControllerState (BatteryChanged, BatteryState, GpsFixChanged, GpsPositionChanged), DeviceState ConnexionChanged, WifiState (auth channel list V2, country, environment), Calibration (EnableMagnetoCalibrationQualityUpdates), Follow_me_Stop. Differences from nap: this mppd no longer names SkyControllerState AttitudeChanged or CalibrationState (Magneto…) descriptors; they moved into libmppd_hw_mpp (strings), together with Skyctrl_SettingsState_CPUID, which mppd_hw_send_all_settings sends once from property ro.serialno when non-empty. GPS descriptors are still referenced but mppd_hw_is_gps_supported is 0 and the module logs “GPS module not supported by this hardware”. Verified.
Local telemetry sinks: libmppblackbox (mux channel, piloting info and button actions to the phone) is unchanged in role. HUD senders (hud_update_ctrl_attitude, _gps, _battery_level, _drone_model, _camera_constraints, _record_state) still exist; the only HUD transport I found is the UDP copy to 127.0.0.1:54321; where the hud_update_* messages go (the pomp context for them) was not decoded and there is no hudd here (Open).
Router: strings “rule found: %s”, “id=0x%08X, mask=0x%08X, proxies=(” and a 4-bit proxy set loop are present at 0x28aca and 0x28b60, matching the nap router. The rule table contents were not decoded here (Open); the nap report’s reading (controller-owned projects consumed locally, other messages forwarded) is Inference for this image too.
Update paths (puf container checks, md5 file, “refuse to downgrade software version”, USB stick, ro.updater.watch.dir, mux updater channel, pinst_setup_trigger) are the same shape as nap with puf_* functions replacing plf_*. No signature check strings were found (control: md5 strings exist). Verified strings.
5. ARSDK in libarsdk (usr/lib/libarsdk.so, 349,528 bytes)
Section titled “5. ARSDK in libarsdk (usr/lib/libarsdk.so, 349,528 bytes)”Revision string libarsdk:c5d7b85f8a85679b0e34901a19c5e66e16a2bc56, equal to etc/manifest.xml drones/arsdk-ng (nap: d0abf562). Verified.
5.1 Device type names (arsdk_device_type_str, 0x2010c)
Section titled “5.1 Device type names (arsdk_device_type_str, 0x2010c)”Switch on id - 0x900, 23 entries via a tbb at 0x20114 with the table at 0x20118; each case loads a pc-relative string (decoded with a script, cases 0x20130 to 0x201a6). Result is identical to the nap library:
| Id | Name | Id | Name |
|---|---|---|---|
| 0x900 | RS | 0x90d | POWERUP |
| 0x901 | BEBOP | 0x90e | EVINRUDE |
| 0x902 | JS | 0x90f | SKYCTRL_2 |
| 0x903 | SKYCTRL | 0x910 | WINGX |
| 0x905 | JS_EVO_LIGHT | 0x911 | PAROS |
| 0x906 | JS_EVO_RACE | 0x913 | SKYCTRL_NG |
| 0x907 | RS_EVO_LIGHT | 0x914 | ANAFI4K |
| 0x909 | RS_EVO_BRICK | 0x915 | SKYCTRL_2P |
| 0x90a | RS_EVO_HYDROFOIL | 0x916 | CHIMERA |
| 0x90b | RS3 | 0x90c | BEBOP_2 |
| 0x904, 0x908, 0x912 | UNKNOWN |
0x90f is SKYCTRL_2: confirmed. Verified. (The strings BEBOP2, RS_TRAVIS, JS_MARSHALL and similar elsewhere in the library belong to the model-name tables, not to this function.)
5.2 Update file and media folder tables (data at 0x4900c and 0x4914c, pairs of id and string pointer, 8 bytes each)
Section titled “5.2 Update file and media folder tables (data at 0x4900c and 0x4914c, pairs of id and string pointer, 8 bytes each)”Media folder (/internal_000/<dir>-style names): 0x901 Bebop_Drone/, 0x90c Bebop_2/, 0x911 Paros/, 0x914 Anafi4k/, 0x916 Chimera/, 0x903 SkyController/, 0x90f and 0x915 SkyController_2/, 0x902 Jumping_Sumo/, 0x905 Jumping_Night/, 0x906 Jumping_Race/, 0x900 Rolling_Spider/, 0x907 Airborne_Nigh/, 0x909 Airborne_Cargo/, 0x90a Hydrofoil/, 0x90d Power_Up/, 0x90e Disco/, 0x910 Swing/. New against nap: 0x90e and 0x910. 0x913 is absent. Verified.
Update file: 0x901 bebopdrone_update.plf, 0x90c bebop2_update.plf, 0x911 paros_update.plf, 0x914 anafi4k_update.tar, 0x916 chimera_update.plf, 0x903 nap_update.plf, 0x90f mpp_update.plf, 0x915 mpp2_update.tar.gz (nap library: mpp2_update.tar), 0x902/0x905/0x906 jumpingsumo_update.plf, 0x900/0x907/0x909/0x90a rollingspider_update.plf, 0x90d powerup_update.plf, 0x90e disco_update.plf, 0x90b and 0x910 rollingspider_update.plf. This image’s own ro.updater.filename is mpp_update.plf which matches the 0x90f row. Verified.
5.3 Projects and commands
Section titled “5.3 Projects and commands”609 g_arsdk_cmd_desc_* dynamic symbols (608 commands plus g_arsdk_cmd_desc_table), nap 603. Counts by project from the symbol names: Common 85, Ardrone3 163, Minidrone 80 (nap 78), Jpsumo 53, Skyctrl 80 (nap 77), Powerup 33, Follow_me 24, Wifi 14, Rc 11, Drone_manager 8, Mapper 16, Mapper_mini 5, Debug 4, Controller_info 2, Thermal_cam 6 (nap 5), Animation 21, Generic 3. Same 17 project names as nap. Added commands: Minidrone_NavigationDataState_{DroneAltitude,DroneQuaternion,DroneSpeed}, Minidrone_RemoteControllerState_ConnectionChanged, Skyctrl_Calibration_{StartCalibration,AbortCalibration}, Skyctrl_CalibrationState_MagnetoCalibrationStateV2, Thermal_cam_Calibration_infos; removed: Minidrone_MediaStreaming_VideoEnable and ..._MediaStreamingState_VideoEnableChanged. Verified (set difference of the two symbol lists).
Skyctrl classes (27, same names as nap): WifiState, Wifi, Device, DeviceState, Settings, SettingsState, Common, CommonState, SkyControllerState, AccessPointSettings, AccessPointSettingsState, Camera, ButtonMappings and State, AxisMappings and State, AxisFilters and State, GamepadInfos and State, CoPiloting and State, Calibration and State, ButtonEvents, Factory, CommonEventState. The legacy ButtonMappings, AxisMappings, AxisFilters and GamepadInfos classes are present and not referenced by mppd; Mapper (project 0x8a, 16 commands, all used) is the mapping interface in practice. Verified for names and non-use.
Other transports compiled in: arsdk_backend_net, arsdk_backend_mux, arsdk_publisher_avahi, arsdk_publisher_mux, arsdk_discovery_avahi/net, FTP/media/updater interfaces. usr/bin/arsdk-cli is the test client. Verified.
6. Bottom board protocol (host side)
Section titled “6. Bottom board protocol (host side)”6.1 USB descriptors in lib/firmware/bottomboard.hex (and bottomboard.dv.hex, identical descriptors at different offsets)
Section titled “6.1 USB descriptors in lib/firmware/bottomboard.hex (and bottomboard.dv.hex, identical descriptors at different offsets)”Intel hex decoded to 18,656 bytes (bottomboard.hex, extents 0x0 to 0x48df) and 18,624 bytes (.dv). Device descriptor at 0x484c: bcdUSB 2.00, class 0, EP0 8 bytes, VID 0x03eb, PID 0x2406, manufacturer string index 1, product index 2, one configuration. Strings: MPP GAMEPAD at 0x487c, www.parrot.com at 0x4888. Configuration (0x4822, 0x29 bytes): bus powered with remote wakeup (0xa0), 100 mA, one interface, class 3, subclass 1, protocol 3, endpoints 0x81 IN and 0x02 OUT, interrupt, 32 bytes, bInterval 10 ms; HID descriptor reports a 0x50 byte report descriptor. Verified. Which of bottomboard.hex or .dv.hex is flashed depends on ro.revision greater than 4 or not (usr/bin/bottom_board_update).
6.2 HID report layout and host protocol
Section titled “6.2 HID report layout and host protocol”Report descriptor (0x47bd, 80 bytes, decoded item by item):
| Report id | Direction | Content |
|---|---|---|
| 1 (Generic Desktop, Joystick) | input | Button page usages 1 to 22, 22 single bits, then 2 constant pad bits (3 bytes), then X, Y, Z, Rx, Ry, 8 bits each, logical -100 to 100. Payload 8 bytes |
| 2 (vendor page 0xff00, usage 1) | input and output | 16 bytes each way, logical 0 to 255 |
Report 1 is consumed by the kernel and surfaces as the evdev device (section 3.5). Nap board had 24 buttons and 8 axes (11 byte payload); this board has 22 and 5 (8 bytes). Verified.
Report 2, board to host, as parsed by libmppbb read callback at 0x130c (read(fd, buf, 0x80), 200 ms watchdog pomp_timer_set re-armed on every frame, “bottom board USB connection : OK/KO” in mppd): the frame is parsed only when the read length is greater than 16 (so 17 bytes: id plus 16), byte 0 is 2 and byte 1 is 0.
| Bytes (hidraw frame) | Meaning | Tag |
|---|---|---|
| 0 | report id 2 | Verified |
| 1 | must be 0 | Verified |
| 2 to 5 | battery voltage in millivolts, little endian u32 | Verified |
| 6 to 9 | flags u32: bit 0 set means CHARGING (callback gets mV 0, percent -1, state 0); bit 1 set means CHARGED (state 1, percent 100, mV as read); the names come from mpp_bb_battery_state_str |
Verified branches, Inference for the physical meaning of the two bits |
| 10 to 13 and 14 to 15 | u32 and signed 16-bit value, passed once to a second callback only when ro.bb.bootloader.version is greater than 4 |
Verified for the parse; Inference that it is board RTC time and a timezone value (the host command that sets them is mpp_bb_rtc_timestamp_set(u32, u16), and mppd has the strings “RTC not initialized, do not use it”, “RTC Timezone not saved, defaulting to GMT”) |
Battery state classification, state numbers from the tbb at 0x1b5c (CHARGING, CHARGED, DISCHARGING, DISCHARGING_LOW, DISCHARGING_CRITICAL, AUTO_SHUTDOWN):
| State | Name | Condition |
|---|---|---|
| 0 | CHARGING | flags bit 0 |
| 1 | CHARGED | flags bit 1 |
| 2 | DISCHARGING | mV above 7400 (0x1ce8) |
| 3 | DISCHARGING_LOW | mV above 7200 (0x1c20), up to 7400 |
| 4 | DISCHARGING_CRITICAL | mV above 6600 (0x19c8), up to 7200 |
| 5 | AUTO_SHUTDOWN | mV 6600 or below |
Percentage: piecewise linear through the five points (7000, 0), (7200, 5), (7400, 15), (7700, 60), (8300, 100) in mV and percent (rodata at 0x1bb4, 8-byte entries; below the first point 0, above the last 100). The callback (mV, percent, state) fires when any of the three changes, except that while the last state was 2 to 5 an update with mV greater than or equal to the last mV is suppressed (voltage is only reported downward during discharge). The voltage range (about 6.6 to 8.4 V) suggests a two-cell pack; the nap’s thresholds (10965, 11065 mV) implied three. Inference. libmppd_hw_mpp (callback at 0x4ab4) mirrors the state into property iosd.charge.device (“1” for states 0 and 1, “0” otherwise, which iosd watches; “Stopping ios charge” is in iosd). Verified for the property writes; the iosd side is Inference from its strings.
Host to board, libmppbb, always write(fd, buf, 0x11) (17 bytes) and success only if 17 was written (0x14f8). Byte 0 is 2 (report id), byte 1 is the command:
| Byte 1 | Function (libmppbb, address) | Payload |
|---|---|---|
| 0 | LED set 0x17b8, LED blink 0x1818 | set: byte 2 = 1 (or 0 when color index 0, off), bytes 3 to 5 = R,G,B from a 10-color table, byte 6 = 0x1f. blink: byte 2 = 4, bytes 3 to 5 = color 1, byte 6 = 0x1f, bytes 7 to 8 = period in ms big endian, byte 9 = the fourth argument (duty/cycle), bytes 10 to 12 = color 2, byte 13 = 0x1f. Colors 1 to 9: red, green, blue, yellow, cyan, magenta, white, (28,6,0) orange, (26,31,16) pale; 0 is off (values 0 to 31, so a 5-bit scale) |
| 1 | per-axis threshold 0x18b4 | byte 2 = axis, byte 3 = threshold |
| 2 | stay on 0x18f2 | byte 2 = flag |
| 3 | RTC timestamp 0x1928 | bytes 2 to 5 = u32 (LE), bytes 6 to 7 = u16 |
| 4 | auto reboot 0x1966 | byte 2 = flag |
| 5 | zero (center) calibration 0x199e | bytes 2 to 6, five values; string form is %d:%d:%d:%d:%d |
| 6 | min/max calibration 0x1a38 | bytes 2 to 11, ten values; string form is ten %d separated by : |
| 7 | raw mode 0x1b04 | byte 2 = 0 or 1 |
At start mppd_hw_create (0x4bf8) calls mpp_bb_new, mpp_bb_start, sets stay-on 1, sets the threshold of axes 0 to 4 to 75 (0x4b, calls at 0x4d1e to 0x4d50), then pushes the factory calibration strings from properties ro.factory.js.centers and ro.factory.js.min_max (filled at early boot from /factory/joystick_zero_value.txt and /factory/joystick_min_max_value.txt by etc/boxinit.rc) through the two _str helpers. Verified. A shell alias in etc/profile.d/reboot runs mpp_bb_cli stayon 1 before /sbin/reboot; mppd_hw_shutdown (0x4ddc) sets auto-reboot or stay-on before reboot(). Verified.
libmppd_hw_mpp capability constants (tiny functions at 0x4bc0 to 0x4bd0): shutdown_on_power_release 0, is_gps_supported 0, has_wifi_ap 0, get_mobile_connectivity 3, get_wifi_bands_supported 1. Nap: has_wifi_ap 1, is_gps_supported 1, wifi bands 3, mobile connectivity 3 (the shutdown value was not recorded there). Verified.
6.3 usr/bin/mpp_bb_cli and companions
Section titled “6.3 usr/bin/mpp_bb_cli and companions”mpp_bb_cli links libmppbb, libmpphw, libpomp. Actions (usage text): on <color>, blink <color1> <color2> <period> <cycle>, th_set <axis> <th> (axis 0 to 4, threshold 1 to 100), th_del <axis>, stayon <flag>, autoreboot <flag>, bat, raw <flag>, debugon, debugoff, quit, -i interactive. Verified. usr/bin/memtester_led.sh drives magenta blinks through it. usr/bin/mppinput BUTTON_NAME|KEYCODE opens /dev/input/event0 and writes a key event (“cannot send key event”); its --list names run A, B, C, X, Y, R1, L1, START, LEFT_JS, RIGHT_JS, L2, R2, VIRTUAL_LEFT_R/L/U/D, VIRTUAL_RIGHT_R/L/U/D, VIRTUAL_SLIDER_R/L, in the same order as the keycode table in 3.5 (Inference for the exact codes, since its table is built in code and was not decoded). Whether a write to that node reaches mppd is Inference (input_inject_event, Recollection of evdev_write).
libmpphw.so (juba discovery for mpp_bb, nap_bb, mpp2_gamepad, plus P7 temperature sysfs reads) and libmppblackbox.so are small and unchanged in role.
7. Binary and library inventory
Section titled “7. Binary and library inventory”usr/bin has 82 regular files and 92 symlinks (all 92 point into bin/busybox, plus iptables-xml to xtables-multi), usr/sbin 40 files and 45 symlinks, usr/lib 87 files and 84 symlinks (soname links and the terminfo, xtables, engines, libnl, wifid directories). BusyBox is v1.25.0 built 2017-10-24 (nap: 2017-08-09). glibc is EGLIBC 2.19, GCC Linaro 4.9-2016.02, kernel modules for 3.4.11+. Origin and version come from etc/manifest.xml, embedded lib<name>:<rev> strings and strings. “Rebuilt” means the file exists in both images with a different md5 (checked for the 22 files I compared, every common Parrot binary differs because the build differs).
7.1 usr/bin
Section titled “7.1 usr/bin”| Path | What it is | vs nap |
|---|---|---|
mppd |
controller hub, 153,716 B, ET_EXEC, rev c3c4fb2d = manifest drones/mpp/mppd |
rebuilt, different hardware layer, no hud/gps link, puf not plf |
aoad |
Android Open Accessory bridge (libaoa), rev f49f43ac | rebuilt |
iosd |
iOS accessory gadget and USB role daemon (g_aea, libusb, PAL/Acid), rev f07c9f18 |
new |
usbnetd |
DHCP/network manager for eth*,usb* (dnsmasq or udhcpc), rev 86cbce70 |
new |
mppinput |
key-event injector for /dev/input/event0, rev 03c2f8bb |
new |
mpp_bb_cli |
bottom board CLI over libmppbb | replaces nap_bb_cli |
gamepad_plugged |
shell: writes 0 to POWER_KEEP when 03eb:2406 appears | new |
bottom_board_update |
shell: dfu-programmer atxmega32c4 flash of bottomboard.hex or .dv.hex, then launch |
changed (two images, no mpp.services.start) |
macgyverd |
crash and oops report daemon (etc/macgyver.conf), rev ea03f58b |
new |
report-generic |
report generator called from boxinit rules | new |
usbctrl-mon |
watches USB controller errors | new |
avabye |
mDNS responder, rev 47732564, INTERFACE=eth*,usb* |
rebuilt, config interface changed |
sensorsd-mpp |
IMU and magnetometer daemon (libiio) | rebuilt |
wifid, wifid-cli |
Wi-Fi manager, STA mode, Broadcom driver plug-in | rebuilt |
arsdk-cli |
ARSDK test client, rev 5c5d5beb | rebuilt |
pomp-cli |
libpomp client | rebuilt |
adbd |
Android debug bridge | rebuilt; started only if ro.debuggable=1 |
crashdump, kmsgd, ulogcat, ulogctl, ulogger, ulogwrapper |
crash dump and log stack | rebuilt |
jubamountd, jubadump |
USB storage automount and juba dump | rebuilt |
pinst_trigger, pinst_version, ubi_check |
installer boot mode, UBI | rebuilt |
shs-cli, cprop, gprop, sprop, wprop |
settings and property tools | rebuilt |
pstart, pstop, pterm, prestart, askfirst |
boxinit service control, console wrapper | rebuilt |
factory_reset |
shell: rm -rf /data/*, run by boxinit when /update/.factory_reset exists (etc/boxinit.rc:76-77) |
present in both (nap script not compared) |
i2ctool, iio_generic_buffer, lsiio |
I2C and IIO tools | rebuilt |
memtester, memtester.sh, memtester_led.sh, radio_test.sh, sensor_test_imu.sh, sensor_test_magn.sh |
factory test helpers (udev rule 70-memtester.rules) |
new |
evtest, libevdev-tweak-device, mouse-dpi-tool, touchpad-edge-detector |
libevdev tools | rebuilt |
lsusb, usb-devices, usbhid-dump, libusb-config, update-usbids.sh |
USB inspection | rebuilt |
dfu-programmer |
Atmel DFU flasher 0.7.2 | rebuilt |
openssl (1.0.1u), c_rehash, htop 1.0.2, top2, ps_procps, vmstat, strace, strace-graph, strace-log-merge, iperf 2.0.5, tree, nsenter-ng, ldd, chattr, lsattr, uuidgen |
third party tools | rebuilt |
udevd, udevadm |
udev | rebuilt |
xmllint, xmlcatalog, xmlwf, xml2-config, ncurses5-config, pcap-config |
libxml2, expat, ncurses, libpcap helpers | rebuilt; nap also shipped pcre-config, pcregrep, pcretest |
| 92 BusyBox symlinks | applets | rebuilt |
| absent here, present in nap | hudd, hdmi_hotplug, nap_bb_cli, nap_factory_reader, gpsd, btgpsdemux, start_dnsmasq_wlan1, pcre* |
7.2 usr/sbin
Section titled “7.2 usr/sbin”| Path | What it is | vs nap |
|---|---|---|
bcmwl, bcmdl |
Broadcom wl control tool and firmware downloader (BCM43526) | new |
dnsmasq |
DHCP server, 2.66 (started by usbnetd) |
rebuilt |
iw |
nl80211 tool | rebuilt |
tcpdump, iftop, genl-ctrl-list, nl-class-*, nl-cls-*, nl-qdisc-*, nl-link-list, nl-pktloc-lookup, nl-classid-lookup |
capture and netlink tools (libnl 3.x soname 200.19) | rebuilt |
xtables-multi and links iptables, ip6tables, -save, -restore |
iptables | rebuilt |
e2fsck, mke2fs, tune2fs, resize2fs, dumpe2fs, e2image, e2undo, e2freefrag, e4defrag, badblocks, blkid, debugfs, filefrag, fsck, fsck-ng, logsave, mklost+found, sfdisk-ng, uuidd and fs links |
e2fsprogs and util-linux-ng | rebuilt |
update-usbids.sh |
script | rebuilt |
| 45 symlinks | BusyBox applets and fs aliases | |
| absent here, present in nap | hostapd, hostapd_cli, wpa_supplicant, wpa_cli, wpa_passphrase |
7.3 usr/lib
Section titled “7.3 usr/lib”| Library | What it is | vs nap |
|---|---|---|
libarsdk.so |
ARSDK core (c5d7b85f) | newer revision, 609 descriptors |
libmppd_hw_mpp.so |
hardware backend of mppd (gamepad, LED, sensors, shutdown), rev c3c4fb2d |
replaces libmppd_hw_nap.so |
libmppbb.so |
bottom board hidraw library, rev c0fe32ba | replaces libnapbb.so |
libmpphw.so |
juba discovery of mpp_bb,nap_bb,mpp2_gamepad plus P7 temperature |
same revision string e3e86c76 as nap (md5 not compared) |
libmppblackbox.so |
blackbox over mux, rev 1f1f5a32 | rebuilt |
libmux.so, libpomp.so, libaoa.so, libjuba*.so (3), libputils.so, libfutils.so, libshs.so, libpinst.so, libulog*.so, libcrashdump.so, libcutils.so, libbase.so, libcrypto_utils.so |
Parrot base libraries | rebuilt |
libpuf.so |
update container (puf) library | new; mppd calls puf_* where the nap mppd did plf header checks |
plib.so |
PLF helper | present in both |
libtar.so |
tar | new |
libnetmon-bcm43526b.so, usr/lib/wifid/bcmdriver.so |
Broadcom network monitor and wifid driver plug-in | new |
acid.so, pal-core.so, pal-devs.so, pal-drivers.so, pal-juba.so, pal-utils.so |
Parrot PAL/Acid runtime used by iosd |
new |
libevdev.so.2.1.12, libiio.so, libconfig.so.9.2.0, libconfig++.so.9.2.0 (mapping parser), libjson-c.so.2.0.2, libcrypto.so.1.0.0, libssl.so.1.0.0 (OpenSSL 1.0.1u), libusb-0.1.so.4.4.4, libusb-1.0.so.0.1.0, libudev.so, libudev-private.so |
third party | rebuilt |
libnl-3, -cli-3, -genl-3, -idiag-3, -nf-3, -route-3 (200.19.0), libpcap.so.1.1.1, libpopt, libproc-3.2.8.so, libxml2.so.2.9.0, libexpat.so.1.5.2, tinyxml.so, libyaml-0.so.2.0.4, libz.so.1.2.11, libncurses/libform/libmenu/libpanel 5.9, libdaemon.so.0.5.0, libnss_mdns* (6), libunwind* (6), liburcu* (7), libip4tc, libip6tc, libiptc, libxtables.so.10.0.0, e2initrd_helper, xml2Conf.sh, engines/* (12 OpenSSL engines) |
third party | rebuilt |
| absent here, present in nap | Qt 5.2.1 (10 libs and qt5/plugins), libMali, libUMP, libEGL*, libGLES*, libion_parrot, libhx270_*, libhud*, libvideo-streaming, libvideo-metadata, librtp, libh264, libgps, libquaternion, libpcre*, libwpa_cli, fonts/ |
8. Differences from nap (one list)
Section titled “8. Differences from nap (one list)”- Identity: 0x090f
SKYCTRL_2,mpp_update.plf, buildmpp-linux-1.0.7(2017-10-24) versus 0x0913SKYCTRL_NG,nap_update.plf,nap-linux-2.0.0-rc3(2017-08-09). 0x0913 is still absent from the media and update tables of this newerlibarsdktoo. - Board: USB 03eb:2406 “MPP GAMEPAD”, 22 buttons and 5 axes, RGB LED through the board, 2S-range battery table. Nap: 2407 “NAP GAMEPAD”, 24 buttons and 8 axes, GPIO LEDs, 3S-range thresholds.
- Mapping: flat directory, four files, no
ro.factory.type, no Z button, no emergency binding, flips bound, slider axis instead of MINI axes, 22 button bits with START, JS and L2/R2 added. - Safety: takeoff refused in battery states 4 and 5 (7200 mV and below) instead of state 2; emergency still has one unguarded call site and no default binding.
- Video and UI: no
hudd, no Qt, no HDMI, no libvideo-*; same SDP relay (TCP 7711, UDP 5004 to peer:55004, SDP address 192.168.53.1). - Radios and sensors: Broadcom STA only (no AP, no
hostapd), no GPS, ICM-20608 and AK8963 sensors,sensors_*insidelibmppd_hw_mpp. - New programs:
iosd,usbnetd,macgyverd,mppinput,usbctrl-mon,report-generic,libpuf, PAL/Acid libraries. - Updater:
pufinstead ofplfchecks. - Router, ARSDK tables, mux channel numbering: same shape as nap, not re-decoded here.
- The nap report’s mux host reading (“skycontroller at 192.168.53.1”) does not match this image: here
mux_add_host("skycontroller", 127.0.0.1)(literal decoded at 0x25b90) andmux_add_host(drone, 192.168.42.1)(0x25ba8); 192.168.53.1 is used only in the SDP. I did not re-check the napmppdfor this, so the nap statement is neither confirmed nor refuted (Open for nap).
9. Not examined
Section titled “9. Not examined”- The board’s own firmware code (only descriptors and strings of
bottomboard.hexwere decoded): which physical key sets which HID button bit, what raw mode and calibration change, what the second report-2 field really is. - The kernel image and HID driver bindings (not in R); the exact evdev codes the unit exposes.
libmuxwire format,libarsdkargument encodings,avabyewire behaviour,wifidandbcmdriver.sointernals,usbnetdinterface-to-subnet rule,iosdprotocol beyond strings.- The router rule table contents and set meanings (0x28aca to 0x28c90).
- The HUD message transport (
hud_update_*), the loopback debug listener port, the local bind address of the 44444 listener. - The LED module’s colour and blink policy (
libmppd_hw_mppled_*), and themppd_hw_save_time/ RTC flow. - Data partitions:
/factory,/data/lib/mppd,/var/lib/mppd. - Qt/GL stack (absent), BusyBox applet config,
etc/inetd.confexposure (belongs to the services note).
10. Open questions
Section titled “10. Open questions”- Does the unit’s
/factoryor/var/lib/mppdhold mapping copies that bind action 26 or a different button set than the defaults? Needs the partitions (read-only file reads). - Which physical keys set HID bits 0 to 6 on the board, and does the board ever send bits 7 to 21 (START, JS, L2/R2, virtual directions)? The relayed list (288 to 294) says no; the board firmware or a long
evtestcapture would settle it. - Does Linux produce codes 704 to 709 for HID buttons 17 to 22 (Recollection of
hid-input.c), making the VIRTUAL_RIGHT and SLIDER codes 720 to 725 unreachable?evteston the unit would show the capability list. - What is
ro.hardwareon a running unit, and isetc/default.mppboard.prop(debuggable, adb TCP 9050, console on ttyPA3) therefore active? The kernel command line orgetpropwould answer. - How does the unit present a USB-NCM link when the image contains no NCM gadget module, and which interface name does
usbnetdthen serve? - What does the second board field (u32 plus s16 at report bytes 10 to 15) carry, and what does
mppddo with it? - What is the router’s rule table for this build, and does the phone’s
Ardrone3_Piloting_Emergencyalways pass it? - Where do the HUD updates and the 127.0.0.1:54321 copy go, now that
huddis absent (dead sink, or a process I have not seen)? - Which address does the 44444 listener bind to, and does
avabyeadvertise on everyeth*/usb*interface (it is configured withINTERFACE=eth*,usb*)? - For the nap
mppd: what do itsmux_add_hostcalls really pass (see section 8, item 10)?
Comments
Spotted a mistake, or tried this on your own Bebop? Leave a note. Comments are kept on this site's own server.