RT-Thread Daily Intelligence Report (2026-11-10)
Research window: 2026-11-09 08:00 ~ 2026-11-10 08:00 (Beijing time, 24 hours)
Sources: RT-Thread official news API and official website news page, GitHub main repository commits / PR / Issue / releases / tags, RT-Thread org repository push list (both the rt-thread and packages repos had pushes this period), five groups of Google News queries in Chinese and English (via proxy), HN Algolia, EETimes China RT-Thread column, club.rt-thread.org forum (still restricted by the site’s WAF this period)
This Period’s Index
- I. Kernel and driver merges
- WCH Qingke kernel fast interrupts: interrupt attribute unification across 27 files (11-09)
- ESP32-C3 linker script: ring buffer IRAM rule missing an asterisk (11-09)
- CherryUSB EHCI: SETUP packet cache not flushed, August old defect closed out (11-09)
- Bouffalo ADC: range validation added to DMA sampling channels (11-09)
- utest: automatic multi-round runs and inter-loop resource sampling (11-09)
- II. Defects and proposals
- RISC-V PLIC: complete before dispatch in non-DM mode (11-09)
- dfs/nfs: server read length was once treated as copy length (11-10)
- ESP32-C3 v1.x: boot header chip version field mismatch (11-10)
- Pending review queue: 140 proposals, 362 issues (11-10)
- III. Official and community events
- Embedded competition concludes: grand prize goes to device-side Agent multi-sensor fusion (11-09)
- Both first prizes in motors: power allocation and dual-core FOC (11-09)
- No new items on the official news API this period: still at the five articles from 11-06 (11-10)
- IV. Ecosystem and vendor adaptation
- software packages: WCH CH32V2/V3 SDK synchronized update (11-09)
- Aerolynq Wi-Fi/Bluetooth host driver enters package index (11-08)
- Community side: energy-storage dual-core learning board and event framework (11-09)
- V. Industry applications and industry-education
- Three implementation lines in the award list: motors, vision, device-side intelligence (11-09)
- Where RISC-V boards stand in the award list (11-09)
- VI. Summary and trend observations
- All five merges are in BSP and driver layers, zero changes to kernel core
- Three ordering issues: external values before local validation
- The RISC-V line is still the same reporter
- Version cadence still at v5.3.0
I. Kernel and Driver Merges
1.1 WCH QingKe Kernel Fast Interrupt: Unifying Interrupt Attributes Across 27 Files (11-09)
Date: 2026-11-09 11:12 (merge time) Source: PR #12004, commit 95e82ec2
27 files, 1087 insertions, 1424 deletions, authored by kaidegit, opened November 7, merged November 9. Different WCH QingKe cores do not support fast interrupts uniformly; cores with only two hardware interrupt levels should not use them, and interrupt declarations across BSPs and drivers had previously each been written their own way. The change adds a CONFIG_WCH_USING_FAST_IRQ configuration option, uses unified interrupt attribute macros to select fast or normal interrupts per chip series, and adjusts the interrupt handlers of peripheral drivers and BSPs one by one; the CH32V203 startup file is switched to the new usage. Validation boards were ch32v208w-r0 and ch32v307v-r1, default configuration.
Deletions outnumber insertions, indicating a consolidation: interrupt declarations copy-pasted across multiple BSPs are pulled back into a single set of macros. The accompanying SDK update lives in the package repository and resolves the issue that had to be worked around in v203; see 4.1.
1.2 ESP32-C3 Linker Script: Ring Buffer IRAM Rule Missing an Asterisk (11-09)
Date: 2026-11-09 11:10 (merge time) Source: PR #12009
One file, one line changed. In sections.ld, the IRAM rule for esp_ringbuf was written as *esp_ringbuf?(...), missing an asterisk before the section list, so it matched no object file; .flash.text in turn excluded that symbol, so the linker placed ringbuf.o’s 0x1fa4 bytes into an orphan .text section after .flash.text. .flash_rodata_dummy reserves space only by SIZEOF(.flash.text), and the orphan section was not counted; once the end of .flash.text was less than 0x1fa4 from the 64KB boundary, code and read-only data landed in the same flash MMU page, rodata was read as code, the serial port printed machine code, and the program failed to start.
The fix is simply adding the missing *. The author provided before/after ranges: before the fix, IROM ran from 0x42000020 to 0x42070f54 and DROM started at 0x3c070020 (both using the 7th MMU page); after the fix, IROM ends at 0x4206f028 with no overlap. This class of problem can also trigger under the default configuration, depending on how much code is added.
1.3 CherryUSB EHCI: SETUP Packet Cache Not Flushed, August Defect Closed (11-09)
Date: 2026-11-09 14:31 (merge time) Source: PR #12008, issue #11749
One file, 4 insertions, authored by boa-z. This is a backport of CherryUSB upstream commit 41f5796: on platforms with non-coherent caches, the EHCI controller could read a stale SETUP packet, causing device enumeration to fail. The CherryUSB copy bundled with RT-Thread is pinned at v1.6.1 and does not include this fix. Issue #11749 was filed on August 25 and is closed in this issue.
Four lines of code close out an enumeration failure spanning three months, exposing a version-tracking problem for bundled third-party components: upstream fixed it, the repo did not follow, and users had to backport it themselves.
1.4 Bouffalo ADC: Range Validation Added for DMA Sampling Channels (11-09)
Date: 2026-11-09 09:29 (merge time) Source: PR #12003, issue #11949
One file, 48 insertions, 2 deletions, authored by Wooohaaa. The Bouffalo ADC driver used the caller-supplied channel number directly to compute the DMA buffer index; negative, out-of-range, and BSP-unconfigured channels would read invalid sample slots. The fix validates the channel’s buffer range and the BSP configuration list before configuring ADC/DMA and waiting on the read semaphore, and replaces the original offset and capacity numbers with named constants.
Validation was host-side only: three configurations (unconfigured channel, CH0 only, CH10 only) were simulated, the configuration table and validation function were extracted and compiled with strict GCC warnings, and all 256 values of rt_int8_t were exhaustively tested; no Bouffalo hardware was used and no full BSP build was performed.
1.5 utest: Automatic Multi-Round Runs and Inter-Round Resource Sampling (11-09)
Date: 2026-11-09 09:31 (merge time) Source: PR #11983
Three files, 123 insertions, 4 deletions, authored by sexenteen, submitted November 3, merged November 9. The change lets the test framework run test cases for multiple consecutive rounds and sample free memory and the thread table (free/ps) between rounds—stress testing shifts from “run once and check the result” to “run multiple rounds and watch the resource curve”; besides the utest component itself, one CI workflow was also modified.
This aligns with several other test-infrastructure items in the queue: dlmodule regression coverage (#11989) and ulog output call sites (#11987) are still under review. Expanding regression coverage first, then adding observability, has been the fixed order on this line since late September.
II. Defects and Proposals
2.1 RISC-V PLIC: Complete Before Dispatch in Non-DM Mode (11-09)
Date: 2026-11-09 22:35 (report) / 23:20 (proposal) Source: Defect #12010, Proposal #12011
While reviewing the ArtInChip D213 BSP (M-mode, device model disabled), reporter boa-z found that in the new RISC-V common port, handler_plic_irq calls plic_irq_ack_or_eoi first, then hands the interrupt to handler_irq for dispatch. With level-triggered sources, the same IRQ can be dispatched repeatedly before the device handler has a chance to clear the interrupt source. The PLIC specification requires the completion action to follow servicing; the DM path already does rt_pic_handle_isr first and plic_irq_eoi after. This code was introduced by #10320, and the old virt64 driver had the same ordering issue fixed by #11305.
#12011, filed 45 minutes later, changes two files with 35 added lines: the handler is moved ahead of completion; at completion time, masked interrupt sources are temporarily enabled, because PLIC ignores completion writes to disabled sources (following the approach of Linux’s SiFive PLIC driver); the shared EOI helper also affects the DM path. Verification was limited to local builds: the qemu-virt64-riscv default DM configuration builds and boots, and the non-DM path used a temporary M-mode BSP not committed to the repo, where the 16550 UART THRE level interrupt was dispatched only once; Actions was not triggered.
2.2 dfs/nfs: Server Read Length Was Used as Copy Length (11-10)
Date: 2026-11-10 01:28 (proposal time) Source: Proposal #12012, Defect #11917
Two files, 41 added lines, 7 deleted lines. nfs_read() previously used the READ3resok.count returned by the server directly as the memcpy length, so an abnormal or malicious server could write past the caller’s buffer. The fix clamps each copy to min(server count, requested count, decoded length), advances the buffer and offset by the actual bytes copied, and bails out on zero-progress replies to avoid spinning; the XDR decode limit is tightened from about 4GB to 1024 bytes (DFS_NFS_MAX_MTU); additionally, WRITE3resok.count is clamped to the submitted length to prevent the server from pushing the write offset beyond the range already sent.
The defect was filed on October 22, with the proposal appearing this issue. This is the most technically significant item this issue: a network filesystem treating the peer’s numbers as local constraints is a classic trust-boundary mismatch.
2.3 ESP32-C3 v1.x: Chip Revision Field Mismatch in Boot Header (11-10)
Date: 2026-11-10 01:36 (proposal time) Source: Proposal #12013, Defect #11991
Two files, 3 added lines, 1 deleted line. The prebuilt bootloader and app images only set the old min_rev=3, so v1.x chips (e.g. rev 1.1) are rejected by esptool and the second-stage bootloader. The fix follows ESP-IDF’s full revision range, changing to min_rev=0, min_rev_full=3, max_rev_full=199, updating the header fields of builtin_imgs/bootloader.bin accordingly and recomputing the trailing SHA-256.
The author notes this is a workaround at the header-field level: there is no CMake in the environment, so the IDF bootloader cannot be rebuilt; no on-device flashing verification was performed either. The longer-term approach remains regenerating the prebuilt images with the current IDF.
2.4 Review Queue: 140 Proposals, 362 Issues (11-10)
Date: 2026-11-10 (verification) Source: Proposal list, Issue list
Down two each from last issue’s 142 proposals and 363 issues, with the net decrease essentially from the five merges this issue. Among the older proposals still progressing within the window, several are worth continued attention:
- APLIC level-interrupt MSI retriggering (#11966): the counter stalls under QEMU with AIA configured; the root cause is an unconditional setipnum write at EOI, generating spurious MSIs even after the interrupt has been withdrawn.
- Terminal tty ioctl user-pointer validation (#11937, reported October 28, still unfixed): affects all RT-Smart boards since v5.1.0, at the level of arbitrary kernel read/write, and remains the highest-risk item in the queue.
- RT-Smart mailbox pseudo-object (#11960): present since v5.0.0, reproduced on QEMU AArch64.
- ivshmem window validation (#11993): adds a fallback switch, off by default, validating size, address overflow, and pointer-representable range before mapping.
- Last issue’s RISC-V FPU context proposal (#12006): adds fs0 through fs11 and fcsr to the switch frame, still under review.
- CAN driver lifecycle refactor (#11779, opened September 2): still updated within the window, not yet merged.
III. Official and Community Activities
3.1 Embedded Contest Concludes: Grand Prize Goes to On-Device Agent Multi-Sensor Fusion (11-09)
Date: 2026-11-09 17:30 (published) Source: 2026 RT-Thread Embedded Contest Rankings Announced
The 2026 RT-Thread Embedded Contest announced its final rankings, with the article published as an original piece by the official RT-Thread WeChat column on EET-China. Prize structure: Grand Prize 3,000 yuan, First Prize 2,000 yuan, Second Prize 1,000 yuan, Third Prize 500 yuan, and Excellence Awards as development board gift packs, all with certificates.
The Grand Prize is “Multi-Sensor Device Information Fusion Monitoring System Based on an On-Device Agent Kernel” on the Edgi-Talk platform: distributed sensor acquisition, MQTT message aggregation, on-device multi-source information fusion and anomaly assessment/decision-making, and an LVGL local dashboard with parameter configuration, placing the Agent’s decision-making on the device side rather than the cloud. This aligns with the on-device AI narrative the official team has repeatedly emphasized this year—pushing inference down while keeping control on the real-time side.
3.2 Both First Prizes in Motors: Power Distribution and Dual-Core FOC (11-09)
Date: 2026-11-09 (published) Source: Original contest results
Both First Prizes are in the motor direction. The first, NS800-PD-FOC on the NS800RT7P65D development board, implements multi-port power conversion across a high-voltage DC port, a low-voltage DC port, and a three-phase brushless motor, combining motor drive, energy scheduling, and a real-time system into one platform—not a PWM or SVPWM example. The second is dual-core sensorless FOC with CANopen, using RT-Thread 5.3.1’s AMP mode on the same board to separate the control core from the communication core, targeting 24V brushless motors.
Second Prize is FeatherTalk, integrating touch interaction, audio recording/playback, Bluetooth, USB audio, filesystem, wireless networking, and local AI on the Infineon PSoC Edge E84 dual-core. The author’s rationale is straightforward: official materials are mostly single-peripheral or middleware examples, lacking a reference project that demonstrates a complete product form.
3.3 No New Official News This Issue: Still the Five from 11-06 (11-10)
Date: 2026-11-10 (verification) Source: Official news list
The API still returns the same five articles from November 6 (elexcon full-stack capabilities, Vinaro industrial PLC salon, Southeast Asia university outreach, Lingang automotive software salon, and the Industry Expo AI car Workshop), with the latest at 11-06 15:54. All were covered in the previous issue and are not repeated here. The news list on the official site is client-rendered; the API is authoritative for content.
IV. Ecosystem and Vendor Adaptation
4.1 software packages: WCH CH32V2/V3 SDK Synchronized Update (11-09)
Date: 2026-11-09 10:27 (merge time) Source: Packages proposal #2052, Main repo proposal #12004
Four files, 18 added lines. This SDK update accompanies #12004: the SDK version numbers and startup file usage for the CH32V2 and CH32V3 series are adjusted together, so the main repo can uniformly configure fast interrupts according to core capabilities. The two repos landed in sequence on the same day (SDK updated at 10:27, main repo merged at 11:12), a typical paired merge in RT-Thread where “driver changes depend on package index.”
4.2 Aerolynq Wi-Fi/Bluetooth Host Drivers Enter Package Index (11-08)
Date: 2026-11-08 18:23 (submission time) Source: Packages proposal #2053
A proposal under review adding a pair of Wi-Fi and Bluetooth host drivers to RT-Thread, targeting Aerolynq combo modules based on Infineon Wi-Fi/Bluetooth silicon, with SDIO plus HCI UART interfaces; the author notes these are proprietary drivers. The value of such host-side wireless drivers lies in board-vendor adaptation: modules can be used directly, without each vendor writing its own protocol-stack glue layer.
4.3 Community Side: Energy-Storage Dual-Core Learning Board and Event Framework (11-09)
Date: 2026-11-09 (push time) Source: VoltDuo, coact
Three pushes within the topic:rt-thread window, two directly related to RT-Thread. VoltDuo is a dual-core controller learning board for energy storage, with a DSP and MCU dividing energy management and real-time control; coact is a C++17 active-object and hierarchical state machine framework, with a single set of headers supporting RT-Thread and other platforms, suited to extracting state machines from hand-written callbacks. The third, zlgopen/awtk, is a high-frequency commit of a general-purpose GUI toolkit with no specific RT-Thread connection, included only as ecosystem reference.
V. Industry Applications and Academia
5.1 Three Deployment Threads in the Winners List: Motors, Vision, On-Device Intelligence (11-09)
Date: 2026-11-09 (published) Source: Original contest results
The three Third Prizes fall into three areas: USB enumeration visualization and descriptor rule diagnostics on ART-Pi2, organizing host requests, device responses, and rule judgments into reviewable session evidence; SENTINEL dual-camera edge AI unmanned-store security, with local storage, no cloud upload, and automatic event reporting; and multi-slave device monitoring and anomaly tracing with DAMO XuanTie D121 as master, using RS485 polling, TF card logging, ESP8266 cloud connectivity, and mini-program remote viewing.
Among the ten Excellence Awards, industrial and consumer split evenly: Fullhan MC6321 smart door lock and MC632X dual-camera anti-tailgating access control, small neural network compensating FOC on NXP MCXA346 (mean speed error in the low-speed steady-state tail reduced by about 4.26%), GD32H759IM USB secondary display plus OV2640 camera two-in-one, ArtInChip D12x portable industrial debugger, and ART-Pi2 master-slave distributed motor servo. The common thread is that none are concept demos—each addresses a clear field problem.
5.2 Where RISC-V Boards Stand in the Winners List (11-09)
Date: 2026-11-09 (published) Source: Original contest results
Two winning entries explicitly run on RISC-V: the Third Prize multi-slave monitoring uses DAMO XuanTie D121 as master, and the Excellence Award portable industrial debugger uses ArtInChip D12x with LVGL v9, packing Modbus-RTU polling, RS485 field device scanning, serial transceiving, analog oscilloscope, and PWM signal source into a handheld terminal.
Both are industrial field tools rather than demos: one addresses retention and tracing of multi-slave data, the other turns “debugging from carrying a laptop to the site” into “power on and use.” Combined with last issue’s RISC-V FPU context defect and this issue’s PLIC ordering defect, it is clear that most RISC-V problems on RT-Thread are not manufactured in QEMU, but discovered in real industrial board M-mode scenarios.
VI. Summary and Trend Observations
6.1 All five merges land in BSP and driver layers, zero changes to kernel core
Date: 2026-11-09 Source: Main repository commit log
The five merges in this window are WCH fast interrupt, ESP32-C3 linker script, CherryUSB EHCI, Bouffalo ADC, and utest. The changes are concentrated in two places, bsp and components, with zero changes to the src directory. The kernel version remains where it has been since September: the official release is still v5.3.0 (September 10), while the mainline self-reports 5.4.0 (see the version field of defect #12010).
For an RTOS that has been around for twenty years, this distribution is normal: new requirements land first in chip vendors’ BSPs, and the kernel only moves when defects accumulate to a certain volume.
6.2 Three ordering problems: external values before local validation
Date: 2026-11-09 to 2026-11-10 Source: Defect #12010, PR #12012, PR #12013
The three new defects in this issue take different forms, but share the same root cause: act first, validate later. PLIC completes the interrupt before dispatching it; NFS copies according to the server-provided length before discussing boundaries; ESP32-C3 hardcodes the version field before discussing compatibility. The direction of the fixes is also consistent—move the judgment before the action, or clamp the value supplied by the peer into the locally allowed range.
Such defects do not surface in functional testing; they appear only under specific trigger conditions: repeated interrupts require a level source to be held continuously, buffer overwrite requires a dishonest server, and version rejection requires a v1.x board. For users, what is worth noting is not any single fix, but that all three occur at the boundary of “interfacing with the outside world.”
6.3 The RISC-V line still comes from the same reporter
Date: 2026-11-09 Source: Defect #12010, PR #12011
The RISC-V FPU context defect from the previous issue and the PLIC ordering defect in this issue come from the same person (boa-z), and both were found on ArtInChip D213 in M-mode with the device model disabled. In both cases, a reproduction was written first, followed by a fix, with intervals of four hours and forty-five minutes respectively.
This shows that verification coverage of the RISC-V common port on the non-DM path remains thin: the DM path has QEMU boards permanently in CI, while the non-DM path relies on individuals running real boards. If RT-Thread is to take on more RISC-V industrial customers, this path needs to enter CI.
6.4 Release cadence still stuck at v5.3.0
Date: 2026-11-10 Source: releases, tags
There are no new additions to releases or tags in this issue; the latest is still v5.3.0 from September 10. The repository totals 12,336 stars and 5,489 forks, with the main repository’s last push at 11-09 11:12. The issue on version management CI and the release process (issue #11956) is still in the queue; it was updated on November 6, with no conclusion yet.
Appendix: Source Verification Table
| Source | Verification Result |
|---|---|
| RT-Thread official news API | Five items, latest 11-06 15:54, zero new items within the window |
| Official website news page | List is client-rendered; API used as authoritative, consistent with API |
| GitHub main repo commits | Five commits within the window, all checked by committer time |
| GitHub PRs and Issues | Five merged, three new proposals and one defect; 140 proposals and 362 issues pending review |
| RT-Thread org repos | 12 repos checked sorted by push; pushes in the rt-thread and packages repos |
| releases and tags | No new items; latest remains v5.3.0 (09-10) |
| Google News Chinese | One contest results article within the three-day window already covered; two others were false matches on automotive content |
| Google News English | Zero valid hits; one false match on a coffee open-source project |
| HN Algolia | All eight candidates were false matches on threads and were excluded |
| EET-China RT-Thread column | One contest results article from 11-09 already covered; the 11-06 project share falls outside the window |
| club.rt-thread.org forum | Returned a WAF block page; could not be verified |
| Community repo search | Three pushes under the topic; two covered as ecosystem observations |
Full Source List
- [1] Main repo commit log — https://github.com/RT-Thread/rt-thread/commits/master
- [2] Proposal #12004 (WCH fast interrupt) — https://github.com/RT-Thread/rt-thread/pull/12004
- [3] Package proposal #2052 (WCH SDK update) — https://github.com/RT-Thread/packages/pull/2052
- [4] Proposal #12009 (ESP32-C3 linker script) — https://github.com/RT-Thread/rt-thread/pull/12009
- [5] Proposal #12008 (CherryUSB EHCI) — https://github.com/RT-Thread/rt-thread/pull/12008
- [6] Defect #11749 (EHCI enumeration failure) — https://github.com/RT-Thread/rt-thread/issues/11749
- [7] Proposal #12003 (Bouffalo ADC channel validation) — https://github.com/RT-Thread/rt-thread/pull/12003
- [8] Defect #11949 (Bouffalo ADC out-of-bounds) — https://github.com/RT-Thread/rt-thread/issues/11949
- [9] Proposal #11983 (utest multi-round runs) — https://github.com/RT-Thread/rt-thread/pull/11983
- [10] Defect #12010 (RISC-V PLIC ordering) — https://github.com/RT-Thread/rt-thread/issues/12010
- [11] Proposal #12011 (PLIC completion ordering fix) — https://github.com/RT-Thread/rt-thread/pull/12011
- [12] Proposal #12012 (dfs/nfs read length) — https://github.com/RT-Thread/rt-thread/pull/12012
- [13] Defect #11917 (nfs_read out-of-bounds write) — https://github.com/RT-Thread/rt-thread/issues/11917
- [14] Proposal #12013 (ESP32-C3 version field) — https://github.com/RT-Thread/rt-thread/pull/12013
- [15] Defect #11991 (ESP32-C3 v1.x rejected) — https://github.com/RT-Thread/rt-thread/issues/11991
- [16] 2026 RT-Thread Embedded Contest rankings announced — https://www.eet-china.com/mp/a537945.html
- [17] Package proposal #2053 (Aerolynq host driver) — https://github.com/RT-Thread/packages/pull/2053
- [18] Community project VoltDuo — https://github.com/lyf-forever/VoltDuo
- [19] Community project coact — https://github.com/DeguiLiu/coact
- [20] RT-Thread official news list — https://www.rt-thread.org/news.html
- [21] Proposal #11966 (APLIC MSI retrigger) — https://github.com/RT-Thread/rt-thread/pull/11966
- [22] Defect #11937 (tty ioctl user pointer) — https://github.com/RT-Thread/rt-thread/issues/11937
- [23] Defect #11960 (RT-Smart mailbox fake object) — https://github.com/RT-Thread/rt-thread/issues/11960
- [24] Proposal #11993 (ivshmem window validation) — https://github.com/RT-Thread/rt-thread/pull/11993
- [25] Proposal #12006 (RISC-V FPU context) — https://github.com/RT-Thread/rt-thread/pull/12006
- [26] Proposal #11779 (CAN driver refactor) — https://github.com/RT-Thread/rt-thread/pull/11779
- [27] Proposal #11989 (dlmodule regression) — https://github.com/RT-Thread/rt-thread/pull/11989
- [28] Proposal #11987 (ulog call sites) — https://github.com/RT-Thread/rt-thread/pull/11987
- [29] releases — https://github.com/RT-Thread/rt-thread/releases
- [30] tags — https://github.com/RT-Thread/rt-thread/tags
- [31] Proposal list — https://github.com/RT-Thread/rt-thread/pulls
- [32] Issue list — https://github.com/RT-Thread/rt-thread/issues