tinyusb.git/src/portable, branch copilot/fix-hil-tinyusb-job

Merge pull request #3775 from hathach/claude/add-usb-debug-sniff

2026-07-18T17:33:14Z

usb-target-debug/usb-sniffer skills

fix(ci_fs): address code-review findings in host/device drivers

2026-07-17T17:18:27Z

Host (hcd_ci_fs.c): - Release the speculatively-armed sibling BDT on the NAK path (IN only) as well as on completion, so a NAKed multi-packet IN no longer leaks a BDT that stays own=1 and blocks every same-direction pipe. Both paths now go through a single release_sibling_bd() helper (was a copy-pasted disarm). - Clear the ENTIRE shared BDT (both directions) on bus reset; clearing only the IN half left a stale OUT/SETUP descriptor after a disconnect mid-OUT, blocking the first control transfer on re-enumeration. - Size bda[] to span the whole BDT (2*2*4) so STAT-indexed access is within the declared array bounds (was out-of-declared-bounds, benign via union). Shared (ci_fs_type.h): - Hoist buffer_descriptor_t and the TOK_PID enum out of the device and host drivers into the shared header so the identical definitions cannot drift. Board (kinetis_k): - Drop a redundant local in board_get_unique_id. Build-verified: host + kinetis k/kl/k32l + MCX. HIL: frdm_k64f host 2/2 (cdc_msc_hid + device_info); frdm_kl25z device core suite green with the relocated definitions.

dcd_lpc17_40: address review findings in the iso paths

2026-07-17T17:15:53Z

From a second max-effort review of the branch: - Drop the dead TUSB_XFER_ISOCHRONOUS case in dcd_edpt_open: iso endpoints are armed via dcd_edpt_iso_alloc/activate (TUP_DCD_EDPT_ISO_ALLOC is defined for this IP), never through dcd_edpt_open, so the case and its dd->isochronous assignment were unreachable and asserted a false invariant. Only bulk/interrupt reach the switch now. - Extend the iso compile gate to the classes that actually arm an iso endpoint: DCD_ISO_ENABLED now includes CFG_TUD_BTH (bth_device.c opens an iso voice endpoint). Without it a BTH build would compile the iso machinery out and fail SET_INTERFACE at runtime. - Un-skip LPC175X_6X in the usbtest example: it shares dcd_lpc17_40.c with LPC40XX verbatim, so the "DCD has no isochronous support" skip reason no longer holds. Build-verified for lpcxpresso1769 (previously blocked by the skip). - TU_ATTR_UNUSED on the ep_id_is_iso helper: every caller is under #if DCD_ISO_ENABLED, so non-iso builds don't reference it and clang's -Wunused-function (fatal in CI) rejected the build — gcc stays quiet. Verified with the full lpc17 and lpc40 example sets under arm-clang. A fifth finding — bounding control_ep_read's PACKET_READY spin with a timeout — was implemented and REVERTED: a naive 100k-iteration bound fires on legitimately-slow control reads and intermittently drops the device (hardware-proven by interleaved A/B testing against the pre-fix binary). The infinite wait is retained; the read is only reached once out_received/ out_queued signal data is present, so the theoretical IRQ-off hang is not reachable in practice. Re-verified on ea4088_quickstart: usbtest 30/30 (repeated) + HIL 14/14.

fix(ci_fs host): release stale sibling BDT on multi-packet completion

2026-07-17T10:26:05Z

hcd_ci_fs shares a single BDT set across all pipes. prepare_packets() speculatively arms the sibling (odd^1) BDT of a multi-packet transfer so it can ping-pong without NAKs. When such a transfer ends early (a short IN packet) or fails, the still-owned sibling was never released, permanently blocking the shared BDT for every other pipe. This deadlocked a 2nd device enumerating behind a hub while another device issued descriptor reads (host/device_info with CDC+MSC): the MSC's control transfers could never acquire the BDT, so it never got Set Address. Release the sibling in process_tokdne()'s completion path, but ONLY for a multi-packet transfer (length > max_packet_size): a single-packet transfer never arms a sibling, so that BDT slot may legitimately belong to another pipe's in-flight transfer and must not be disturbed (doing so unconditionally corrupts concurrent transfers, e.g. the CDC bulk-IN vs MSC enum in host/cdc_msc_hid). Mirrors the equivalent device-side fix in dcd_ci_fs.c; the host needs the multi-packet guard because its BDT set is shared across pipes. Verified on frdm_k64f (HIL): host/device_info now enumerates both CDC+MSC behind a hub, host/cdc_msc_hid still mounts the MSC (no regression). Co-Authored-By: Claude Opus 4.8 (1M context) Claude-Session: https://claude.ai/code/session_01ExGPLP5eU43LR7o6yYLpNi

dcd_ci_fs: disarm sibling BDT on short-packet OUT completion

2026-07-17T10:26:03Z

A multi-packet OUT transfer speculatively arms both even/odd BDTs to avoid NAK. When the host ends the transfer early with a short packet, the sibling BDT was left armed (own=1), desyncing the even/odd ping-pong so the next OUT packet landed at buffer+max_packet_size instead of buffer and the stack read stale data. Disarm the sibling on completion. Fixes device/mtp on Kinetis (GetDeviceInfo command was received into the wrong buffer half -> hang). Pre-existing (MSC only arms single-packet command receives so it never hit the double-buffer path). HIL: frdm_kl25z & frdm_k64f device 13/13. Co-Authored-By: Claude Opus 4.8 (1M context) Claude-Session: https://claude.ai/code/session_01ExGPLP5eU43LR7o6yYLpNi

migrate NXP Kinetis khci to chipidea ci_fs driver (device + host)

2026-07-17T10:26:02Z

Complete the khci -> chipidea ci_fs migration that was started for device (commit d70403f1f "host is not yet"): - device: switch kinetis_k/kl/k32l (Makefiles + k32l CMake) to dcd_ci_fs.c - host: add hcd_ci_fs.c (port of hcd_khci.c onto ci_fs_regs_t) and switch all Kinetis families to it; remove src/portable/nxp/khci entirely - enable host examples (device_info, cdc_msc_hid) for mcu:KINETIS_K - README: merge the KL and K32L2 rows into a single "KL, K32L" ci_fs row hcd_ci_fs.c also fixes two pre-existing host bugs found via HIL on frdm_k64f (present in the old hcd_khci.c too): - data toggle was flipped on a NAK in suspend_transfer; a NAK transfers no data so the toggle must be preserved, else the retried bulk packet is silently discarded by the device (MSC CBW/CSW hang). See comment in file. - prepare_packets asserted and dropped a transfer when the single shared BDT was still owned by an in-flight transfer under concurrent activity; now it returns busy and resume_transfer defers/retries on the next SOF. HIL verified on frdm_k64f: device 13/13, host cdc_msc_hid (CDC mount + echo + MSC mount, through a hub). Co-Authored-By: Claude Opus 4.8 (1M context) Claude-Session: https://claude.ai/code/session_01ExGPLP5eU43LR7o6yYLpNi

dcd_lpc17_40: mask USB IRQ around non-reentrant SIE and realization sequences

2026-07-17T09:48:02Z

The SIE command protocol (CmdCode + CCEMPTY/CDFULL handshake), the slave-mode Ctrl/RxData/TxData window, the EpIntEn read-modify-writes, and set_ep_size's ReEp/EP_RLZED handshake are all shared between thread-mode API calls and dcd_int_handler, and none are reentrant: an ISR preempting a thread-mode sequence consumes its handshake flags or, in set_ep_size's case, a bus reset's DevIntClr = 0xFFFFFFFF eats the EP_RLZED flag the spin waits on, hanging it forever. Guard them by masking only the USB IRQ (nestable, ISR-safe; CMSIS NVIC_DisableIRQ already ends with DSB+ISB). control_xact keeps the mask across its in_isr=true event push, since osal_none skips queue locking for in_isr. Hardening, not a fix for an observed failure: the ea4088 usbtest 30/30 + HIL 14/14 results were reproduced with and without it. The windows are a few instructions wide and most exposed on RTOS builds where class drivers queue transfers from tasks concurrent with the USB IRQ.

dcd_lpc17_40: fix stale EP0 out_received, add isochronous support

2026-07-17T09:48:01Z

EP0 control-OUT fix (usbtest 14/21, errno 110/-74): usbd queues the status-stage OUT ZLP of every control read with buffer=NULL, so the ISR's `if (out_buffer)` check missed it and marked the arriving ZLP as out_received instead. The stale flag poisoned the next control-OUT with data: its first chunk "completed" instantly from an empty EP0 buffer and the host's real DATA NAKed forever. Track queued transfers with an explicit out_queued flag and void half-finished control state on a new SETUP. Isochronous support (UM10562 12.15.6): 5-word DMA descriptors with per-packet size memory, buflen/present_count in packets, one packet per FRAME (no DMARSet/EpIntEn involvement), completion at EOT for both directions. Details that matter: - the iso machinery (5th DD word + packet-size memory) is compiled only when an iso-capable class is enabled (CFG_TUD_AUDIO/VIDEO/VENDOR), so non-iso builds pay nothing: _dcd stays 648 B vs 1032 B with iso - ISR dispatch keys on the hardware's fixed ep-number/type map (ep_id_is_iso), never on dd fields that thread mode rebuilds - iso OUT honors Packet_valid (bit 16) and prefills the hardware writeback slots with 0, so a missed frame counts as 0 bytes instead of reading back stale buffer contents as data - packet count is validated (tu_div_ceil <= ISO_MAX_PACKETS) before the DD is touched, so an oversized transfer is refused without leaving a serviceable half-built descriptor armed for the frame engine - dcd_edpt_iso_alloc and iso_activate both enforce the fixed iso endpoint numbers (3/6/9/12); classes ignore alloc's return value, so activate must not trust it Un-skip LPC40XX in the usbtest example; tier 4 now enumerates and passes iso cases 15/16/22/23. cdc_msc_throughput and printer_to_cdc had bulk on iso-only EP3 (SET_CONFIGURATION failed with -32); add the LPC17/40 EPNUM block (bulk on EP2/EP5) like other fixed-EP examples. Verified on ea4088_quickstart: usbtest tier-4 battery 30/30 repeatedly and the full device HIL suite 14/14 (incl. audio_test iso).

dwc2: fix EP0 OUT dcache invalidate range; run usbtest on espressif s3/p4 and mimxrt1015

2026-07-14T19:31:14Z

edpt_schedule_packets() advanced xfer->buffer past each armed EP0 chunk, so the OUT-complete handler invalidated the cache at the ADVANCED pointer: one line past the received data. The CPU then read stale cached bytes instead of the DMA'd packet, and the misplaced invalidate discarded a dirty line of whatever variable follows the buffer - random neighbor corruption on every control-OUT data stage. Found by usbtest ctrl_out (cases 14/21) on espressif_p4_function_ev with DMA enabled, the first DWC2 target combining buffer DMA with a data cache: usbd control state wedged after the first control write (every later request stalled), and one build layout panicked in the usbd memcpy with a wild pointer. Rework the EP0 chunk bookkeeping so xfer->buffer always points at the un-consumed position: the arm no longer advances it; instead the EP0 re-arm paths advance past each completed (full) chunk, invalidating it first on the OUT side. The final OUT completion invalidates exactly the received bytes of its last chunk, taken from DOEPDMA ("incremented on every AHB transaction", databook 7.1.83 - the same semantics the SETUP path relies on) before dma_setup_prepare() re-targets it. EP0 chunking state (ep0_pending) is now also dropped on bus reset and on a new SETUP, so a stale latched completion can no longer re-arm EP0 DMA from dead state. No behavior change for targets without dcache. While root-causing, the FIFO layout was cross-checked against the DWC2 databook/programming guide v4.20a: the existing GDFIFOCFG programming (EPInfoBaseAddr = otg_dfifo_depth - 2*ep_count, one SPRAM word per endpoint direction for buffer DMA) is conformant and needs no change; the P4 HS instance's reset GDFIFOCFG (0x03800400) merely reflects a scatter/gather-sized EP_LOC_CNT of 128 that buffer DMA does not need. With the fix in place, enable the usbtest battery on the espressif fleet: tools/build.py allowlists device/usbtest (a plain IDF component like board_test/video_capture) and both espressif boards' only-lists gain device/usbtest. Also re-enable device/usbtest on mimxrt1015_evk: its skip predated the dcd_ci_hs stale-ACTIVE-overlay fix (already on this branch), which cured the battery that previously killed the uPD720201 host controller twice (2026-07-11 ROM fw, 2026-07-13 case 27 on fw 2.0.2.6); rig-validated 30/30 three consecutive runs. Validated on rig (all 30/30): espressif_p4_function_ev(-DMA) (was 22/30 under DMA), espressif_s3_devkitm(-DMA), stm32f723disco(-DMA), mimxrt1015_evk; p4/s3 slave-mode unaffected (DMA-only code path); compile-checked stm32h743nucleo +TUD DMA, stm32f407disco, stm32l476disco (device ports currently on the dead hub). Co-Authored-By: Claude Fable 5 Claude-Session: https://claude.ai/code/session_017TQZrFfU3K4Y198aLsUpBC

dcd(ci_hs): stale overlay fix; run usbtest on lpcxpresso43s67

2026-07-14T08:23:07Z

- dcd_edpt_stall flushes the primed buffer (ENDPTFLUSH), but the aborted transfer's dQH overlay can be left ACTIVE with mid-transfer state; the next prime after clear-halt then resumes the stale overlay instead of loading the fresh qtd, so post-halt IN reads return mid-buffer data (usbtest case 13 'buf[32] = 56 (not 0)', with case 18 failing downstream of the same corruption in the full battery). qhd_start_xfer now clears overlay.active alongside overlay.halted before linking the new qtd. - test/hil(hfp): drop lpcxpresso43s67's device/usbtest skip - the historical first-case wedge no longer reproduces on this branch, and with the overlay fix the board runs 30/30 on its Fresco xHCI host (previously 28/30 with deterministic case 13/18 failures). mimxrt1064_evk (imxrt dcache path) 30/30 regression-clean. - docs(hil skill): document the external hifiphile rig - pool test/hil/hfp.json, SSH-reachable from htpc/ci with no outbound SSH, exercised by the CI hil-tinyusb (hfp.json) job; never run HIL against it during development unless the user explicitly asks. Co-Authored-By: Claude Fable 5 Claude-Session: https://claude.ai/code/session_017TQZrFfU3K4Y198aLsUpBC