Problem
With PR #3672/#3674 landed, CEP handshakes reliably but the SPI transport
hangs PM5 hard (needs a physical reset, no watchdog on this platform)
after some number of attach/reconnect cycles - sometimes the first
cycle, sometimes dozens of pings in. Logic-analyzer capture of a hang in
progress showed a ~1.2s burst of rapid CS-gated retries, and in one
capture, three periodic transactions after that burst got progressively
slower (2.1ms -> 3.9ms -> 3.2ms -> 5.2ms) instead of recovering.
Root causes (two, compounding)
1. cep_spi_write_sync() had seven bare `while (flag == RESET);` spins
waiting on SPI1's TDBE/BF flags, with no timeout. PM5 is the SPI
slave - TDBE only clears when the Flipper is actively clocking, on
its own ~20-120ms schedule, completely async to when PM5's reply is
ready. Queue a reply between two master polls and the wait never
returns - full main-loop death.
2. Artery's own errata sheet for the AT32F435/437 (ES0003) documents
that in SPI slave hardware-CS mode, the internal clock sync is not
guaranteed to reset on every CS falling edge - a bad sync can persist
across transactions with no self-correction. This is PM5's exact CEP
configuration (SPI_CS_HARDWARE_MODE, slave mode).
Change
armsrc/pm5_cep.c:
- cep_spi_write_sync() rewritten around a bounded cep_spi_wait_tdbe()
helper (same spin-iteration-count pattern already used for reads) -
returns PM3_EIO on timeout instead of hanging.
- New cep_spi_resync(): hardware-resets SPI1 via crm_periph_reset()
(a register re-init alone doesn't clear the errata's bad internal
state) and brings it back up. This is Artery's own documented recovery
for ES0003 ("reset SPI and restart communication"), triggered by our
existing NG-frame CRC/length checks instead of the SPI peripheral's
hardware CRC feature.
- cep_do_handshake()'s SPI-stall recovery path was itself unbounded
(retried forever as long as the CC controller reported "still
attached") and missing the StartTicks()/I2C_init(true) guard every
other I2C call site in this file has (same bug class as #3672) - now
bounded (CEP_HANDSHAKE_REPLY_MAX_RETRIES) with the guard added.
- cep_spi_data_available() now discards the Flipper's 0x55 idle-poll
filler bytes before reading the real length header, instead of ever
interpreting one as (part of) a frame length.
armsrc/cmd.c: wire cep_spi_resync() into every CEP-reachable error exit
in receive_ng_internal() (short/overflow/CRC-mismatch reads) and both
reply_old()/reply_ng_internal()'s write-failure paths, so any detected
corruption resets SPI1 before the next transaction rather than letting a
bad sync persist. Add a WITH_CEP-gated inline stub for cep_spi_resync()
so receive_ng_internal()'s call sites don't need per-site #ifdef guards
on non-CEP platforms.
Behaviour change
None for USB/FPC/BWM. CEP: a corrupted or stalled SPI1 transaction now
self-heals (one dropped reply, then normal operation) instead of
wedging the device.
Testing
- make clean && make PLATFORM=PM5 PLATFORM_EXTRAS=BWM and
make clean && make PLATFORM=PM3RDV4 - both zero warnings.
- Hardware: flashed to a real PM5 paired with a real Flipper Zero FAP.
- 60s continuous-connection capture with zero USB activity (isolates
CEP from any client-side interaction): clean, no anomalies.
- 110s capture with 18 real `hw status` calls interspersed: all 18
succeeded, CEP stayed attached throughout.
- 5 full attach/detach cycles with `hf search` against a real NTAG213
card before/during/after each attach (240s, logic-analyzer capture
throughout): all 15 reads returned the identical UID regardless of
CEP state, zero hangs, final liveness check succeeded with no reset
needed.
- Not tested: clang toolchain (firmware-only diff).
Checklist
- [x] Proper style of own code, no unrelated reformatting included
- [x] Builds warning-free with gcc (PM5 and RDV4)
- [ ] clang not tried (firmware-only diff)
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The two g_tx_completed spin loops in usb_write() had no timeout at all -
if the host ever stopped pulling an IN packet while the device still read
as enumerated, PM5 would spin forever with no recovery path. Add a
spin-iteration bound (not tick-based: g_tx_completed is set from the USB
core's own ISR, and the shared ticks timer is routinely stopped by LF/HF
code at arbitrary points), matching the existing bounded-wait pattern
used elsewhere in this codebase.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
receive_ng_internal() zeroed rx->data (up to PM3_CMD_DATA_SIZE bytes)
between the preamble read and the payload read. CEP's SPI1 receiver has
no FIFO (single-byte holding register), so that stall lost bytes on every
frame - confirmed via logic analyzer (wire content byte-perfect) plus
Dbprintf instrumentation showing the payload read consistently short by
half. Reordered so preamble/payload/postamble/CRC are all read back-to-back
before any memset/memcpy happens. USB/FPC/BWM are unaffected: same total
work, just resequenced.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Adds A1856618 to t55xx_default_pwds. This is the verified working login
password (as used with -p); the bit-reversed on-air form 1866A185 does not
authenticate.
cep_attach_poll() (and cep_init()) call I2C_BufferReadRaw()/I2C_init(true),
which spin on the shared ticks timer via WaitUS()/WaitMS(). That timer is
routinely stopped with StopTicks() by LF/HF protocol code once it's done
with precision timing (em4x50.c, legicrf.c, hitag_common.c, lfsampling.c,
etc). cep_attach_poll() runs unconditionally every 50ms in the main loop
with no idea whether the timer is currently running - if its poll lands in
a window where something just called StopTicks(), WaitTicks()'s
`while (GetTicks() < target)` spins forever on a frozen counter, hanging
the entire main loop irrecoverably. `hf search` cycles through many such
probes, giving this poll many chances to land in that window, matching the
reported "hf search hangs the PM5, needs a battery pull" regression - and
since the poll runs regardless of Flipper attachment, merely compiling in
WITH_CEP was enough to trigger it.
bwm_lowbatt_poll() (armsrc/bwm_charger.c) is the same shape of periodic
main-loop I2C poller and already guards against exactly this by calling
StartTicks(); I2C_init(true); itself immediately before every I2C access.
Mirror that convention in cep_attach_poll() and cep_init().
Re-enables WITH_CEP as the PM5 default (reverts f88880fc6's temporary
disable) now that the underlying bug is fixed. SKIP_CEP=1 remains available
for anyone who wants to opt out at build time regardless.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
doegox asked on PR #3668 whether CEP really needs to be guarded by a
compile flag, suggesting it ship on by default with a SKIP_CEP opt-out
instead - mirrors the existing NO_LOWBATT_SHUTDOWN/NO_LOWBATT_BEEP
opt-out pattern in common_arm/Makefile.hal. WITH_CEP is now baked into
PM5's base PLATFORM_DEFS; PLATFORM_EXTRAS=SKIP_CEP filters it back out.
No change to armsrc/Makefile (already keys off the WITH_CEP define, not
the old extras token) or to the CEP transport logic itself. Updated the
PM5-FLIPPER-USAGE doc, the CHANGELOG entry, and one stale comment in
include/pm3_cmd.h that referenced the old opt-in token.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Pass the requested system code to firmware selection so default and system-specific reader modes share one polling path. Keep the RF field active across retries for the system-specific mode.
Co-Authored-By: Codex (GPT-6) <noreply@openai.com>
Add --sys to the FeliCa reader so repeated polling can activate an Express Transit card while keeping the RF field on.
Co-Authored-By: Codex (GPT-6 Sol) <noreply@openai.com>
Co-Authored-By: Codex (GPT-6) <noreply@openai.com>
Wire up the previously-dead Flipper Zero handshake/SPI code from
at32_unit_test.c into a real WITH_CEP platform extra, independent of
BWM (new g_reply_via_cep flag, not reusing BWM's g_reply_via_fpc, so
both transports can run concurrently).
Fixes a real reconnect bug found during hardware bring-up: the
handshake listener was gated on a CC-controller attach edge that
didn't reliably fire on every FAP relaunch, requiring a physical
cable replug to reconnect. Now watches for a fresh incoming byte any
time the link reports attached, with a bounded receive timeout so a
stray byte can never hang the main loop.
RF/FPGA-dependent functionality (hf/lf reads) is out of scope here -
tracked separately. Closes#3667.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
lf t55xx dump: when lf t55xx detect has already recovered and confirmed
the password, T55xxReadBlockEx now trusts that session config instead of
re-probing the config block unauthenticated. A password-gated tag does
not answer that probe, so 'dump -p' failed its safety check on every
block and demanded --override; it now reads directly.
lf t55xx p1detect / lf search: the page-1 trace fingerprint hardcoded
Atmel's manufacturer byte (0x15/0x39) into a 16-bit preamble, so a valid
T5577 clone with another manufacturer (e.g. Tendyron 0x65) was rejected.
The duplicated atmel/silicon checks are collapsed into
t55xx_trace_preamble_match(), which also matches the bare 8-bit ACL 0xE0
- the same anchor lf t55xx trace decodes against.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The module ships with BLE open: anyone in range can connect and run
commands, and there was no way to turn the radio off or to require
pairing from the PM5, although the module already implements bonding
with a static passkey (LE Secure Connections + MITM, SPP characteristic
encrypted) behind commands nothing exposed.
New CMD_PM5_BWM_BLE (0x0183): one action byte, then a full status
snapshot back (bwm_ble_status_t: switch, state, bonding, passkey, TX
power, address, name, bonded devices). Client menu `hw bwm ble`:
status everything above, with a warning while open
on | off persisted radio switch (needs the companion
Proxmark5_BWM_esp32 PR; off =
nothing can connect, USB/WiFi only)
pairing on|off [-k] require the 6-digit passkey, set the passkey; a
change of on/off restarts the stack (drops a
connected client), the factory key 123456 is
flagged
forget -i N | --all remove bonded devices
txpower -a/-c dBm advertising / connection power, -24..18 and 20
The firmware replies before restarting the stack so the reply survives
when the command itself arrived over BLE. Fields a module without the
switch cannot answer read 0xFF and print as unknown. A module that does
not answer at all ends the status snapshot after its first query instead
of running into the client's timeout.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A forward frame on the BWM link is anywhere from ~30 bytes to ~2.1 KB (an
NG reply at PM3_FPC_MAX_DATA), so a 16-frame ack window allowed ~33 KB
in flight while the module held un-acked frames in a 4 KB UART ring:
any radio stall of ~45 ms at 921600 baud overflowed it and the module
dropped PM3 bytes. A frame cap tight enough for big frames stalled
uploads, where the module defers small acks behind large incoming
chunks (the old comment on BWM_FC_WINDOW).
Count bytes instead: a FIFO of un-acked frame lengths (the module acks
in order) with budget BWM_FC_BYTES = BWM_ESP_UART_RX_BUF (12288, the
module's ring in the companion Proxmark5_BWM_esp32 PR) minus 1 KB slack,
and a 64-frame FIFO cap for tiny frames. Change the ring size in both
places; the armsrc -pedantic C99 build rejects a static assert tying
them.
The ack timeout no longer forgets the whole window. That made the gate
open loop for the rest of the command: the acks of forgotten frames
released younger ones early, so after the first eaten ack (bwm_cmd()
drains the ring, #3648) the budget protected nothing. Now it forgets
just enough of the oldest frames to send this one: a lost ack costs one
wait and heals, and a stalled radio gets one frame per timeout on top of
a full ring instead of a window. The wait is 200 ms instead of 50: a
full window that stays full is the module stuck on the radio (a phone
scanning stalls BLE for a few hundred ms), and every timeout leaks a
frame into its ring. A CMD_ERROR broadcast naming SEND_FORWARD_DATA
releases the slot like the ack (older module firmware answers that way
when a WiFi forwarding type is configured and no channel takes the
frame). hw status at debug level prints the window counters.
Measured over BLE with the companion module firmware, which acks a
frame as it leaves its ring: 86 KB/s honest against 88 KB/s open loop,
and a 20 s run with the phone scanning meanwhile finishes clean.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The USB-unplug power-off left two ways to drain the battery: a PM5
woken by the button and forgotten, and one left idle after the phone
disconnected.
Idle trigger (opt-in, PM5_AUTOOFF_IDLE_MS 0 by default so out of the box
only the unplug trigger exists, as before): on battery the board powers
off after the idle timeout with no interaction and no wireless client.
Interaction is a received command over any transport, any button press,
or a client connecting or leaving; a standalone mode or a long-running
command holds the main loop, so neither can be cut short. The BWM
reports its clients with the LINK_STATE broadcast (8092, companion
Proxmark5_BWM_esp32 PR) which bwm_forward.c tracks; at boot the BLE
half is seeded with a status query since the module only reports
changes. When the timer is due but a client is still tracked, the module
is asked for its BLE state (at most every 10 s), so a lost "client left"
broadcast cannot pin the board on. And right before going off the module
is asked once more: older module firmware never sends the broadcast and
a "connected" one can get lost, and neither may power the board off
under a silent BLE client. No answer means off.
--unplug off: the unplug trigger always won, so "survive the unplug, go
off after N idle seconds" was not reachable. It is its own setting
(default on = unchanged). With it off an unplug only restarts the idle
clock - from the unplug, not from the last command, or a board that sat
idle on USB for an hour would still die with the cable. Unplug detection
is edge-based (VUSB present at the previous poll, absent at this one)
instead of "absent and seen since boot", which would fire on every poll
after a tolerated unplug. VUSB is followed while the feature is switched
off too, so switching it on again on battery is not taken for an unplug.
Stored on the BWM: runtime-only settings reset at every boot, which is
exactly when the idle trigger matters. The PM5 has no settings store, so
the switch, the idle timeout and the unplug switch live in the module's
host value slots (APP_CMD_SET/GET_SYS_HOST_VALUE 1021/1022, companion
PR; slot 0 switch, 1 idle seconds, 2 unplug), loaded at the first
auto-off poll and saved by CMD_PM5_BWM_AUTOOFF with 1 s per write, so a
silent or older module still leaves the reply inside the client's 5 s
wait. Without a module, or with an older one, the defaults apply and the
status says "not stored".
hw bwm autooff [on|off] [--idle <sec>] [--unplug <on|off>]; no argument
shows the state, one fact per line: switch, idle timeout, unplug
power-off, stored on module, USB power, USB seen since boot, unplug
trigger armed, tracked client, the module's BLE state, idle time.
Payload [action][enabled][idle_s u32][unplug, optional] with keep
sentinels; both actions are non-zero so an old firmware, which reads
byte 0 as the enable flag, can only be left enabled. hw status prints
the module's live BLE state and the auto power-off setting; at debug
level also the tracked client, which should agree with the live state.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
New `hw bwm wifipower [off|none|min|max]`. `off` turns the BWM's WiFi
fully off (BLE-only, persisted; the same path as `hw bwm wifi stop`,
and the module's default state). The other three set the ESP's persisted
WiFi modem power-save type (CMD_PM5_BWM_WIFI_PS 0x0182, module commands
2052/2053, esp_wifi_set_ps()) used while WiFi is up: min (the ESP-IDF
default, the modem sleeps between DTIM beacons), max (sleeps for the
whole listen interval, lowest current, slower to react), none (modem
always on, lowest latency). With no argument it reports whether WiFi is
up at all and the stored type. Independent of `hw bwm powersave`.
The link helper's "cmd failed" diagnostic is now printed only at debug
level: a status query while WiFi is off fails by design and was
cluttering every call.
Verified on hardware: default reads min with WiFi off; none, max and
min each set and read back; off tears WiFi down; an unknown value is
rejected; max survives a module reboot. Not measured with WiFi
associated.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
New `hw bwm powersave [on|off]` (CMD_PM5_BWM_POWERSAVE 0x0181) shows or
sets the ESP's persisted power-save switch (companion Proxmark5_BWM_esp32
PR: DFS, light sleep and slow advertising after a 30 s fast phase, or
the stock always-on behaviour), so the two can be A/B measured and the
saving turned off when chasing a link problem. `hw status` prints the
state next to the BWM firmware version.
Both module queries in `hw status` are time-bounded (version 2 x 800 ms,
power save 300 ms): a module that is silent, or a firmware without the
command, cannot push the report past the client's timeout.
Verified on hardware: state read/set/persist across an ESP reboot,
`hw status` after 4, 15 and 40 s idle in power-save mode and after 6 s
idle with it off.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Port the "reduce idle power" feature from the Fantasi firmware
(soeinova/Fantasi @ 9671309, plus its later ACC-SOF fix) to the PM5
(AT32F435) target.
Problem
-------
The PM5 firmware ran the ARM core at 288 MHz with a 1.3 V LDO and a
main loop that busy-polled at full speed even when the device was idle.
The 288 MHz PLL and the 1.3 V LDO dominate idle draw, and the busy loop
adds to it. Nothing scaled back when there was no work.
Change
------
New armsrc/pm5_power.c drives a refcounted clock/voltage governor:
- Idle (nothing pending, boost count 0, boot grace elapsed, FPGA off):
take SCLK off the PLL onto HICK-48, power the PLL down, drop the LDO
to 1.1 V, park the FPGA 24 MHz clock (PA8 low), and halt the core in
WFI until the next IRQ or the 1 ms SysTick wake.
- Any timing-critical work holds a 288 MHz boost: AppMain wraps
PacketReceived (which on the PM5 also covers a standalone mode, since
it is only entered through a command there); the BWM low-batt poll
and the RGB indicator wrap their I2C. Boost is refcounted and
IRQ-guarded.
- USB (OS image) moves to the crystal-less HICK-48 clock, ACC-trimmed
against the SOF, so powering the PLL down never touches USB. The
bootrom keeps the HEXT/PLL USB path (AS_BOOTROM), so DFU flashing is
unchanged.
- bwm_uart_clock_update() re-inits UART4 through the SDK usart_init()
on each clock switch (the divider set at 288 MHz is 6x off at 48 MHz).
- FpgaIsOff() (fpga_core.c, PM5 only) tracks the last conf word so the
governor never downclocks while a reader field / emulation is running.
- Wake preamble on the BWM link: the companion module firmware
(Proxmark5_BWM_esp32) light-sleeps 2 s after the last byte on the link
and its UART wakes on RX edges but loses the bytes that carried them.
bwm_uart_write() therefore leads with four 0x55 bytes and a 10 ms
settle on its first write and after 1 s without a write of its own
(RX is not tracked: the ring can be drained long after the bytes
came in). The module only starts light-sleeping once it has seen a
preamble, so the one sent while it was still booting is repeated
after the boot-time link negotiation. A module firmware without light
sleep drops the preamble as noise.
New `hw powersave on|off` toggles the governor at runtime (default on);
`hw status` reports the idle-clock and WFI-asleep fractions.
Behaviour change
----------------
- PM5 idles at 48 MHz with the PLL off and the CPU halted; it boosts to
288 MHz for commands and BWM housekeeping. Command timing and
throughput are unchanged (intra-command work is always boosted).
- PM5 USB now runs off HICK-48 (ACC-trimmed) instead of the HEXT PLL.
- New command id CMD_PM5_POWERSAVE (0x0180).
Testing
-------
Built warning-free (gcc) for PLATFORM=PM5 (with and without
PLATFORM_EXTRAS=BWM), PM3RDV4 and PM3GENERIC, armsrc and bootrom. Client
built via the default Makefile. Not built with clang or on
Windows/macOS; nothing outside PM5/AT32 guards changes.
Flashed to a PM5 and exercised against a MIFARE Classic 4K card:
hw version/status/ping (4000 & 512 B), hw tune, hf search, hf 14a
info/reader (x12 across idle gaps), hf mf info/chk/rdbl, lf search/read,
hw powersave off/on, held-field (raw -sk) correctly blocks the downclock,
hw reset + re-enumerate. All pass; USB stays up across downclocks. Idle
reaches ~80-95% at 48 MHz and ~90%+ WFI-halted after a few seconds idle.
Measured at the USB port: idle draw 283 mA -> 207 mA with the stock
BWM firmware.
Co-authored-by: Soei Nova <soeinova@proton.me>
Co-authored-by: noproto <noproto@zeroday.engineering>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`make commands` on current master: the two hf ntag424 commands had not
made it into the generated docs, and `hw bwm help` is gated by IfBwm
since 339db44cf but the docs still list it as offline-capable. No other
change. Kept separate so the commits that follow only carry their own
doc changes.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Corrects the claim that every block reads back as 00000000 after finalize.
On the tag tested, block 0x38 held 11223344 before finalize and block 0x20
held it afterwards, a shift of 0x18, while blocks 0x10/0x11 and 0x14/0x15
were cleared.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The help text and the -y warning said finalize erases the whole tag memory
to 00000000. A dump of a tag before and after finalize shows otherwise: the
configuration blocks are cleared, but user data is relocated rather than
erased. Data written to block 0x38 before finalize read back at block 0x20
afterwards.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
When no sector key was known, autopwn probed for a static encrypted
nonce using the default key FFFFFFFFFFFF. On a FM11RF08S with
non-default keys that auth fails, the probe returns NONCE_FAIL, and
autopwn fell through to darkside/nested instead of the sen recovery.
The later re-detections are gated on known_key, so nothing recovered it.
Probe the known Fudan backdoor keys first, as 'hf mf info' and
'hf mf sen' already do, and run the detection with MF_KEY_BD. The
default key stays as the fallback for cards without a backdoor.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
