Skip to content

v0.7.2

Choose a tag to compare

@github-actions github-actions released this 14 Nov 05:16
· 10 commits to main since this release

Changed

  • Firmware: frames to the host are queued, not written in place. firmware/common/tx.c encodes each
    frame into one of two rings (one for the USB side, one for the receive side) and feeds the UART
    from them; hal_uart_write is replaced by hal_uart_tx_start, hal_uart_tx_busy and
    hal_uart_tx_wait_done. On the RP2040 and RP2350 the UART output is a DMA channel and the
    critical section around every write is gone: before, an EP_OUT frame, an event or a control data
    stage written by the USB core spun on the UART byte by byte with interrupts off on that core, so the
    target's IN transfers waited for it. The other ports send from the CPU inside hal_uart_tx_start as
    before. Each port's mimusb_limits.h states MIMUSB_TX_RING_BYTES and MIMUSB_TX_RX_RING_BYTES
    (the build stops when they are missing; tests missing_*_limit). The ring replaces the 2 KiB
    long-frame buffer of the USB side.
  • Firmware: an ASK rule on a host-to-device request with a data stage no longer stops the USB side while
    the host decides. The vendor callback of the data stage returns at once and the board keeps the
    status stage the stack asks for until the verdict is in (mimusb_control_hold_status); reports on
    the other endpoints go out meanwhile. mimusb_control_set_single_context is gone, since nothing
    waits any more. The wire protocol is unchanged.
  • Firmware: the queue's lock is released with an atomic exchange, not a store. On the RP2040 (Cortex-M0+) the
    exchange runs under a hardware spinlock and a plain store does not, so a release landing inside the other
    core's exchange was overwritten and the lock stayed taken (the board stopped answering).

Added

  • firmware/tests/test_tx.c: the frame queue against a UART that stays busy (ring wrap, frames of every
    length, two producers in order, a full ring, the rate change after the queue, a producer that does not
    wait for the line).
  • Tests of the held status stage in firmware/tests/test_core.c: refused, accepted, no verdict in time,
    and voided by a new setup packet.

Firmware Binary Third-Party Notices

The host library (libmimusb) is licensed under the MIT License and has no third-party runtime library dependencies.

Prebuilt microcontroller firmware images (mimusb_firmware-*) incorporate third-party SDK and USB stack components under permissive licenses:

Firmware asset Third-party components & licenses
mimusb_firmware-rp2040-*, mimusb_firmware-rp2350-* TinyUSB 0.18.0 (MIT), Raspberry Pi Pico SDK 2.2.0 (BSD-3-Clause, Copyright (c) 2020 Raspberry Pi (Trading) Ltd.), Arm CMSIS Core (Apache-2.0), Newlib C runtime
mimusb_firmware-nrf52840-* TinyUSB 0.18.0 (MIT), Nordic Semiconductor nrfx (BSD-3-Clause, Copyright (c) 2017-2024 Nordic Semiconductor ASA), Arm CMSIS 5 (Apache-2.0), Newlib C runtime
mimusb_firmware-stm32f4-* TinyUSB 0.18.0 (MIT), STMicroelectronics cmsis_device_f4 & stm32f4xx_hal_driver (BSD-3-Clause / Apache-2.0, Copyright (c) 2016-2019 STMicroelectronics), Arm CMSIS 5 (Apache-2.0), Newlib C runtime
mimusb_firmware-esp32s3-* TinyUSB 0.18.0 (MIT), Espressif ESP-IDF v5.4.2 (Apache-2.0, Copyright 2015-2025 Espressif Systems), Amazon FreeRTOS Kernel (MIT), Newlib C runtime

Complete copyright notices, license texts, and warranty disclaimers for all firmware binaries are provided in the release asset mimusb_firmware-THIRD_PARTY_NOTICES-0.7.2.txt and in THIRD_PARTY_NOTICES.md / NOTICE.