RT-Thread Daily Intelligence Report (2026-11-16)
Research window: 2026-11-13 08:00 ~ 2026-11-16 08:00 (Beijing time, 72 hours, covering Saturday and Sunday)
Sources: RT-Thread official news API (club.rt-thread.org, category rt_org), GitHub main repository commits / PR / Issue / releases / tags, all 35 repositories in the RT-Thread org checked in push order, smart-build build repository, topic:rt-thread active repository search, Google News Chinese and English queries (via proxy, including three-day and seven-day windows), Hacker News Algolia
Index for This Issue
- I. Two RISC-V kernel changes
- Floating-point registers: a state leak across context switches (11-13)
- Device tree into procfs: an extra OFW tree under /proc (11-13)
- II. Twenty commits in twenty minutes on Friday morning
- Three proposals flagged for review in the previous issue landed the same day (11-13)
- Partition table and sector table: boundaries of two table indexes (11-13)
- User-space boundaries: serial configuration, legacy USB host, Smart terminal, and the rt-link receive side (11-13)
- Test reinforcement: QEMU regression for dlmodule and atomic operation test case reset (11-13)
- III. Three BSP commits: WCH low power, K230 pulse width, ESP32-C3 older chip
- CH32V208: RISC-V board package gains power management (11-14)
- K230: pulse width readback and a new set of test cases (11-13)
- ESP32-C3: v1.x chips rejected by prebuilt images (11-13)
- IV. Under review and builds
- ArtInChip D213 initial support: same origin as the floating-point fix (11-14)
- QEMU virt configuration synced to 5.4.0 (11-14)
- V. Official channels, community, and ecosystem
- Official news API has no new items for the sixth consecutive issue (11-16)
- No substantive hits in the Google News three-day window (11-16)
- CAN framework refactor: first version closed the same day, reopened the next day (11-14)
- rt-carplay: a CarPlay receiver grown from a D213 board (11-15)
- VI. Summary and trend observations
- The audit line begins to gather evidence
- On the same morning, neither kernel change was a security fix
- Thirty-eight hours with zero commits over the weekend
- Queue decline and one correction of scope
I. Two RISC-V Kernel Changes
1.1 Floating-Point Registers: A State Leak Across Context Switches (11-13)
Date: 2026-11-13 09:37 (merge time) Source: PR #12006, Issue #12005
This one did not start in the kernel repo, but in a port for another board. On November 7, the reporter was porting ArtInChip D213 (C906, M-mode) and found that fs1 held the wrong value after a thread switch. To rule out factors from his own BSP, he went back to the official bsp/qemu-virt64-riscv to reproduce, documenting the environment as QEMU 10.0.13, OpenSBI 1.6, RV64D, S-mode, single hart, and wrote a test case that checks floating-point registers across switch types.
Two problems were confirmed. The first is in the voluntary switch path: when rt_thread_yield switches between two threads both using floating point, the callee-saved fs0 through fs11 are not saved at all, so one thread reads values actually written by the other. The report gives a measured comparison: expected 0x3ff0000100000000, got 0x3ff0000000000000. The second is in the timer preemption path: f0 through f31 are saved, but fcsr is not, so the floating-point rounding mode and exception flags are silently altered—expected 0x68, got 0x24. The former directly corrupts floating-point results; the latter neither crashes nor reports an error, it just lets computed results drift over time.
The PR followed on November 8 and was merged on the morning of November 13, with the issue closed the same day. The change spans five files, adding 117 lines and deleting 7: context_gcc.S gains the save/restore, asm.h, context_ext.h, and stack.h gain offset and layout macros, csr.h gains accessors for fcsr, and it handles stack alignment, vector offsets, and register usage under RV32E, while also initializing state for the first switch.
Worth noting is the verification standard: the fix itself contains no test case. The UTest that reproduces the problem exists only in the issue body and was not merged with the fix. For a defect that silently changes numerical results, this kind of gap costs more than it would for a security fix, because nothing will sound the alarm on the next regression.
1.2 Device Tree into procfs: An OFW Tree Appears Under /proc (11-13)
Date: 2026-11-13 09:37 (merge time) Source: PR #12000, Build Repo PR #51
The other change from the same morning is not about fixing correctness but about adding observability. In RT-Thread, the device tree (OFW) could previously only be queried through C interfaces; the actual running topology was invisible from the command line. This change exports OFW nodes and properties to procfs, adding proc_ofw.c at 189 lines, with six files totaling 218 lines added and 3 deleted. The work includes node and property traversal, segmented reads of binary properties, handling of node name conflicts, and cleanup on initialization failure.
The companion action is in the build repo: smart-build PR #51 enables the option for qemu-virt the same morning. Together, the point is straightforward—after boot, the device tree is visible under /proc, corresponding to the location of /proc/device-tree on the Linux side. For anyone working on BSPs, this is far faster than reading source code.
One detail reveals the maturity of this line of work: the config option lives in components/dfs/Kconfig, is off by default, and is enabled per board. New capabilities come in first without changing the behavior of existing images—this has been the standard approach for such changes this year.
II. Twenty Commits in Twenty Minutes on Friday Morning
Within the window, the main repository saw 24 commits across 73 files, with 2,615 lines added and 217 lines deleted. Of these, 20 landed within a twenty-minute span from 09:27 to 09:47 on the morning of November 13, two more at 16:42 and 18:48 that day, and the final two on the evening of November 14. The author distribution is highly concentrated: 17 commits from the same auditor, 2 from the maintainer himself (the two kernel changes from the previous section), 2 each from two others, and 1 from one more.
2.1 Three Proposals Flagged for Review Last Issue Land the Same Day (11-13)
Date: 2026-11-13 09:32 to 09:46 (merge time) Source: Proposal #12017, Proposal #12019, Proposal #12021
The previous issue’s assessment explicitly listed three proposals “still in the queue”: command-line file operation path construction (CVE-2024-25390), variable export table population (CVE-2024-25392), and YMODEM path copying (CVE-2024-25394). All three merged this issue, within less than a quarter-hour of each other.
The command-line one modifies msh.c, adding 58 lines and deleting 18, changing path concatenation from “concatenate first, validate later” to a construction with a length cap, so that completion paths for commands like ls, cd, and mkdir no longer rely on the caller’s diligence. The variable export table one modifies var_export.c, adding 44 lines and deleting 26, confirming remaining table entry capacity before populating the export table. The YMODEM one modifies ry_sy.c, adding 46 lines and deleting 10, adding a copy boundary to each of the ry and sy paths.
What the three share remains the same: confining length and index arithmetic at the entry point. The difference is that these are no longer new discoveries but positions reached on last month’s list. Last issue noted that “failing review may just be scheduling”; this issue delivers on that guess.
2.2 Partition Table and Sector Table: Boundaries of Two Table Indexes (11-13)
Date: 2026-11-13 09:46 (merge time) Source: Proposal #12034, Proposal #12038
The GPT partition table one modifies efi.c, adding 50 lines and deleting 18. The size of the partition entry table comes from a header field on disk; after multiplying by the entry length there is no upper-bound check, so a crafted image can make this product wrap around or exceed the buffer, and subsequent iteration over entries writes past the table. The fix first confirms the product is representable using division, then confirms the entire table fits, and only proceeds with parsing when both conditions hold.
The CFI flash one is another face of the same problem: the sector table is populated according to the sector count returned by a command, and the index upper bound uses “less than or equal to” rather than “less than,” so the last sector writes one cell past the array. The change is small, and the problem shape is entirely consistent with this series since September: external input determines the loop upper bound.
2.3 User-Space Boundaries: Serial Configuration, Legacy USB Host, Smart Terminal, and rt-link Receive Side (11-13)
Date: 2026-11-13 09:38 to 09:47 (merge time) Source: Proposal #12037, Proposal #12036, Proposal #12031, Proposal #12039
All four occur at the seam between the kernel and the outside world. The serial one copies the RT_DEVICE_CTRL_CONFIG control structure entirely from user space into a kernel copy before parsing; previously it interpreted a user pointer directly as a kernel struct, so any field misalignment became an out-of-bounds read. The legacy USB host stack one validates the device descriptor’s bLength before deciding how much to read; defect #12026 was already covered in the November 11 report, and this issue lands the fix.
The rt-link one modifies the receive side; rtlink.c adds 79 lines and deletes 4, and long-frame reassembly first converges on the allocated buffer length before copying. Both send and receive paths now have boundaries: the send side merged on November 12, the receive side this issue. Defect #12030 closed the same day.
The Smart terminal one modifies tty_wait_background, adding a null check before fetching the process control block. This kind of null pointer does not appear on the normal exit path, only surfacing when a session is destroyed concurrently—a shape only stress testing can hit.
2.4 Test Hardening: dlmodule QEMU Regression and Atomic Operation Case Reset (11-13)
Date: 2026-11-13 09:30 to 09:37 (merge time) Source: Proposal #11989, Proposal #12033
The most substantial commit in this batch is not a fix but a test. The dynamic module loader’s QEMU regression, proposed on November 5 and merged today, spans more than a week and lands at a scale of 19 files and 799 added lines: 325 lines for the test cases themselves, 108 lines of build fixture scripts, 94 lines of CI-side run scripts and 82 lines of validation scripts, plus one integration point in .github/workflows and a utest configuration. Coverage per the proposal description is 10 test cases, including relocatable objects, shared objects, symbol import/export, and repeated load and unload.
The atomic operation case one modifies inter-run cleanup: when running the same suite over multiple rounds, worker threads and semaphores left over from the previous round are not released, and the next round reports false failures due to unclean resources. The change reclaims both at round boundaries.
Together the two show this audit line’s center of gravity is shifting. The main theme of the past two weeks was “find out-of-bounds and fix it”; this issue begins to show investment in “making the next one automatically discoverable.” For a development line preparing for 5.4, that is worth more than fixing five more out-of-bounds bugs.
III. BSP Trio: WCH Low Power, K230 Pulse Width, ESP32-C3 Legacy Chips
3.1 CH32V208: Power Management Added to the RISC-V Board Support Package (11-14)
Date: 2026-11-14 17:28 (merge time) Source: PR #12007
WCH’s CH32V208 is a RISC-V chip built on the Qingke V4F core, and this board support package previously had no power management component. This change adds 637 lines of drv_pm.c, with eight files totaling 725 insertions and 1 deletion. The work covers sleep and wake entry points, peripheral clock gating and restoration, wake source configuration, board-level Kconfig options, plus a readme with power consumption curves and additional CI configuration.
The PR was submitted by a community account on the chip vendor’s side, opened on November 8 and merged on November 14, depending on another low-level change merged earlier the same day. For low-power scenarios, whether a board support package has a PM component determines whether the chip can go into battery-powered products, not just whether it can boot. Closing this gap on RISC-V small cores is more practical than chasing higher clock speeds.
3.2 K230: Pulse Width Readback and a New Set of Test Cases (11-13)
Date: 2026-11-13 09:36 (merge time) Source: PR #12032
K230 is Canaan’s RISC-V vision chip. This change fixes pulse width reading: the driver got the counting direction backwards when reading back duty cycle, so channels configured for up-counting returned inverted pulse widths. The driver itself only changes 11 lines with 6 deletions, but the accompanying test cases add 120 lines and delete 42, covering boundary values, zero duty cycle, and full duty cycle.
Expanding test cases alongside a bug fix is rare among recent BSP changes. Most board-level fixes only touch the driver itself, with verification done by hand. This change flips that ratio, and the author is worth noting: it is the same account that submitted the dlmodule test cases in the same batch.
3.3 ESP32-C3: v1.x Chips Rejected by Prebuilt Images (11-13)
Date: 2026-11-13 16:42 (merge time) Source: PR #12013, Issue #11991
ESP32-C3 belongs to the RISC-V C series, unlike the Xtensa cores of the S2 and S3, so this issue does not involve other cores. The bug report is specific: boards with chip revision v1.1 and ROM version esp32c3-eco7-20230720 refuse to run after flashing the official prebuilt bootloader and application images, while the same images work fine on newer chips.
The cause is that the chip revision field in the image header was filled in for the newer revision, so older chips fail the check at boot stage. The fix restores the revision field to a compatible range so prebuilt artifacts cover v1.x as well. The trouble with this class of problem is that victims cannot work around it themselves: if you have an older chip, you have to build the toolchain and images from scratch. For a board that wins on low price and wireless capability, the coverage of usable images is the entry barrier.
IV. Under Review and Builds
4.1 Initial support for ArtInChip D213: Same origin as that floating-point fix (11-14)
Date: 2026-11-14 18:46 (proposal time) Source: Proposal #12058
D213ECV-DEMO-V4-0 is an ArtInChip evaluation board, built around the T-Head C906, RV64GC, with 128 MiB DDR3. The proposal submits 30 files and adds 1,838 lines, comprising a complete board support package: a 129-line linker script, a 222-line rtconfig.h, a 242-line UART driver, plus four port files for clock, interrupt, system, and timer, along with a bilingual (Chinese/English) board readme.
Two technical points in the approach are worth noting. First, it reuses the RISC-V common port’s M-mode startup, context switching, floating-point state, and PLIC handling; the board level only implements platform initialization and serial adaptation. Second, an external loader completes DDR, PLL, and memory attribute configuration before jumping, meaning this board cannot boot the system directly from internal ROM.
The submitter is the reporter of the floating-point defect from Section 1.1. Connecting the two, the path is clear: to port this board, they first found the floating-point context problem in the kernel, fixed it, and then came back to submit the board package. The same community developer both reporting a kernel defect and delivering a BSP within two weeks is a combination that has appeared more often in RT-Thread this year than in previous years.
4.2 QEMU virt configs synced to 5.4.0 (11-14)
Date: 2026-11-14 21:36 to 21:51 (commit and merge time) Source: Build repo commit 21b1293c, Build repo proposal #52
This commit in the build repo syncs the kernel default configs of qemu-virt-aarch64 and qemu-virt-riscv64 to RT-Thread 5.4.0: it updates the version number, adds newly introduced config symbols, and disables unused Ethernet, ATA/NVME, TEE, and PCI options; on the riscv64 side it additionally enables three capabilities — hook lists, hardware atomic operations, and the CPU-level find-first-set instruction. Corresponding assertions were added to test_qemu.py, and the commit message states the verification result as all 156 pytest items passing.
The version number needs some context: the official release is still v5.3.0 from September 10, while the main branch header has reported 5.4.0 for some time; this commit is the first time the build side writes “5.4.0” into the default configs and aligns to it. Enabling hardware atomics and find-first-set on riscv64 indicates that this version’s baseline capability assumptions for RV64 are being raised, and scenarios still using older cores should watch for config compatibility.
V. Official Statements, Community, and Ecosystem
5.1 Official news API sees no new items for the sixth consecutive period (11-16)
Date: 2026-11-16 (verification time) Source: RT-Thread official news list
The API still returns the same five articles from November 6, the latest timestamped 15:54: RealSyde’s full-stack capabilities and industrial intelligence, the Vinaro industrial PLC tech salon, Southeast Asian university industry-education integration, the Lingang automotive software ecosystem salon, and the AI cart at the Industry Expo. This batch was fully captured on November 9 (a three-day window covering the weekend), and there have been no new items for six consecutive periods since.
The news list on the official site is client-rendered and cannot be scraped, so the API is authoritative. The external communications gap has now lasted ten days; in terms of cadence this falls within a normal intra-quarter lull and should not be read as a signal.
5.2 Google News three-day window yields no substantive hits (11-16)
Date: 2026-11-16 (verification time) Source: Google News search
A Chinese exact-match query over a seven-day window returned three results, all from the same tech site: two on full-stack smart home development (11-12, 11-14) and one on Android evolving into an AI-native platform (11-12). All mention RT-Thread in passing and do not constitute reportable events. Another query hit coverage of the “Xinguang” product launch at the Wuzhen Summit on November 3, which is old news outside the window. English queries returned zero hits in the seven-day window. Bare queries in the three-day window returned normally, ruling out proxy or search-channel issues.
On Hacker News, all nine candidates were literal thread mismatches (thread pools, SIMT, forum posts) and were filtered out by title semantics; none relate to real-time operating systems. This channel has produced no valid items since September.
5.3 CAN framework refactor: first version closed same day, reopened the next (11-14)
Date: 2026-11-13 14:44 to 11-16 03:05 (proposal and updates) Source: Proposal #12054, first-version proposal #12053
The CAN framework refactor went through a full “withdraw and resubmit” cycle this period. The first version, #12053, was filed on the afternoon of November 13 with 28 files, 7,357 insertions, and 2,851 deletions, and was closed by the author an hour and a half later. It was reopened the next day at 17:19 as #12054, expanded to 42 files, 9,187 insertions, and 5,292 deletions, with three linked issues: the framework refactor plan, transmit-side full refresh and hardware abort capability, and driver lifecycle and transmit handling.
The proposal text states the responsibility boundary as a criterion: the generic CAN framework handles only hardware resource management, synchronization, ownership, terminals, and lifecycle; transmit buffering, retries, scheduling, and protocol QoS policies go to the upper protocol stack or application. This is a refactor that draws the boundary clearly rather than a local patch to a single BSP, so its scale and review cycle will exceed the norm. It was still under review at this period’s cutoff, with the most recent update in the early hours of November 16.
5.4 rt-carplay: a CarPlay receiver grown from a single D213 board (11-15)
Date: 2026-11-14 22:36 to 11-16 07:18 (repo creation and pushes) Source: boa-z/rt-carplay, boa-z/luban-lite
A newly created community repository positioned as a portable CarPlay receiver SDK, with ports for both Linux hosts and RT-Thread. Its structural description is clearer than most personal projects: the protocol core and receiver runtime can be compiled with no dependency on RT-Thread, Luban-Lite, ArtInChip, or GUI headers, while platform services, identity and pairing, media backends, and input ownership are provided by the host through public C interfaces.
Timeline: the repository was created in the early morning of November 15, and ten consecutive commits were pushed in the early hours of November 16 (04:27 to 07:18), focused on isolating the vehicle status message handler, the UI sync session, and the fallback video channel. Meanwhile the Luban-Lite side was pinned to the v1.3.3 snapshot, with D213 board assembly and board testing kept in Luban-Lite while the SDK itself remains portable. The author notes that lifecycle audits and bilingual documentation audits are still in progress, and that historical D213 impressions cannot serve as hardware validation for the current refactor.
An ecosystem note in passing: AWTK had five commits on November 13, focused on text layout and undo-record merging. These are general GUI improvements with no specific RT-Thread connection, noted here only for ecosystem observation.
VI. Summary and Trend Observations
6.1 The audit line begins to add evidence
On the numbers alone, the audit line still dominates this period: 17 of the 24 commits come from the same author, and the vast majority are of the “external input determines loop bound” type. But the structure is changing. The dynamic module regression tests proposed on November 5 were merged this period with 19 files and 799 lines, of which 325 lines are actually executed test cases and nearly 300 lines are fixtures and CI scripts; the K230 pulse-width fix also expanded its test cases from 42 to 120 lines.
This shows the line is filling a gap that has long been missing: making the next occurrence detectable automatically. The main theme from September to October was clearing the backlog; from this period onward, “tests before or alongside the fix” pairings are appearing. For a security line that has been patched for three straight weeks, this is a signal of shifting from manual labor to infrastructure.
6.2 Two kernel changes on the same morning, neither a security fix
The two commits at 09:37 on the morning of November 13 sit adjacent to each other, both by maintainers themselves, covering the RISC-V floating-point context and OFW export via procfs. Their nature is entirely different from that batch of 20 audit fixes: one fixes a correctness defect that silently changes numerical results, the other adds runtime observability.
The juxtaposition is worth noting. RISC-V’s weight within RT-Thread is shifting from “it runs” to “it runs correctly and visibly”: saving floating-point state across context switches is the prerequisite for running correctly, and being able to see the device tree from the command line is the prerequisite for visibility. The defect was found by the community while porting a new board, the maintainer produced a proposal within four hours, and it was merged six days later—a chain that has appeared several times this year and is currently the healthiest link, its only weakness being that fixes come without test cases.
6.3 Thirty-eight hours with zero commits over the weekend
The time distribution is clean: 22 commits on Friday, 2 on Saturday evening (CH32V208 and sys_poll count wraparound), and then zero commits to the main repository for 38 hours from 17:29 on November 14 to 08:00 on November 16. All of November 15 saw no commits, no new issues, and no new proposals.
The two Saturday commits were long-queued merges rather than new development, indicating the maintenance window extended into the weekend. Actual development activity ended in those twenty minutes on Friday morning. This “Friday burst, weekend silence” pattern has been the norm rather than the exception over the past month.
6.4 Queue decline and one correction
The pending queue dropped from 163 proposals and 358 issues last period to 144 and 342. Within the window, 24 were merged and 7 newly created (including the first-version CAN refactor closed the same day), while 16 issues were closed and 0 newly created. This is the first time in nearly three weeks that proposals and issues have declined simultaneously.
One timing correction to the previous period is also needed. The previous period stated that the pair of four commits for M0 and M23 were merged on November 12; after checking GitHub merge records, the actual merge time for the four commits was 09:30 to 18:48 on the morning of November 13 (the author’s commit times were indeed November 12 to 13). These four were covered last period and are not recounted here; this note only clarifies the difference in source and timing.
Appendix: Source Verification Table
| Source | Verification Result |
|---|---|
| RT-Thread official news API | Five items, latest 11-06 15:54, sixth consecutive issue with zero new items |
| Official website news page | List is client-rendered; API is authoritative and consistent with the API |
| Main repo commits | 24 commits, 73 files, 2615 lines added, 217 lines deleted; 22 on Friday, 2 on Saturday |
| PRs merged within window | 24, one-to-one with commit count |
| PRs opened within window | 7, of which the first version of the CAN refactor was closed the same day |
| Issues | 16 closed, 0 opened |
| Pending queue | 144 PRs, 342 issues (previous issue: 163 / 358) |
| RT-Thread org repos | 35 checked sorted by push; only rt-thread and smart-build had pushes within the window |
| packages repo | No pushes; last at 11-09 10:27 |
| releases and tags | No additions; latest remains v5.3.0 (09-10); main branch header self-reports 5.4.0 |
| topic:rt-thread active repos | 4, of which three are general-purpose embedded projects and were excluded |
| Google News Chinese | Within the seven-day window: three tech-site roundups and one old item outside the window, none substantive |
| Google News English | Zero hits within the seven-day window; bare queries work normally, ruling out channel failure |
| HN Algolia | All nine candidates were literal “thread” false matches and were excluded |
| club.rt-thread.org category API | rt_news category does not exist; rt_org returns five items normally |
Full Source List
- [1] PR #12006 (RISC-V floating-point state save) — https://github.com/RT-Thread/rt-thread/pull/12006
- [2] Bug #12005 (QEMU virt64 floating-point state anomaly across context switches) — https://github.com/RT-Thread/rt-thread/issues/12005
- [3] Commit 9d670b73 (preserve FPU state) — https://github.com/RT-Thread/rt-thread/commit/9d670b739c72123e90380995f77ec6a2c3f0977f
- [4] PR #12000 (OFW exported via procfs) — https://github.com/RT-Thread/rt-thread/pull/12000
- [5] Commit 548fb6c3 (procfs OFW) — https://github.com/RT-Thread/rt-thread/commit/548fb6c356f76379a687d2939cfbc8921ae371fe
- [6] Build repo PR #51 (enable OFW procfs on QEMU virt) — https://github.com/RT-Thread/smart-build/pull/51
- [7] PR #12017 (command-line path construction) — https://github.com/RT-Thread/rt-thread/pull/12017
- [8] PR #12019 (variable export table capacity) — https://github.com/RT-Thread/rt-thread/pull/12019
- [9] PR #12021 (YMODEM path copy) — https://github.com/RT-Thread/rt-thread/pull/12021
- [10] PR #12034 (GPT entry table overflow) — https://github.com/RT-Thread/rt-thread/pull/12034
- [11] PR #12038 (CFI sector table out-of-bounds) — https://github.com/RT-Thread/rt-thread/pull/12038
- [12] PR #12037 (serial config user-space integer conversion) — https://github.com/RT-Thread/rt-thread/pull/12037
- [13] PR #12036 (USB host descriptor length validation) — https://github.com/RT-Thread/rt-thread/pull/12036
- [14] Bug #12026 (legacy USB host descriptor out-of-bounds write) — https://github.com/RT-Thread/rt-thread/issues/12026
- [15] PR #12031 (rt-link long-frame reassembly boundary) — https://github.com/RT-Thread/rt-thread/pull/12031
- [16] Bug #12030 (rt-link reassembly trusts peer size) — https://github.com/RT-Thread/rt-thread/issues/12030
- [17] PR #12039 (Smart terminal null pointer) — https://github.com/RT-Thread/rt-thread/pull/12039
- [18] PR #11989 (dlmodule QEMU regression) — https://github.com/RT-Thread/rt-thread/pull/11989
- [19] PR #12033 (atomic operation test case resource reset) — https://github.com/RT-Thread/rt-thread/pull/12033
- [20] PR #12007 (CH32V208 power management) — https://github.com/RT-Thread/rt-thread/pull/12007
- [21] PR #12032 (K230 pulse-width readback) — https://github.com/RT-Thread/rt-thread/pull/12032
- [22] PR #12013 (ESP32-C3 version field compatibility) — https://github.com/RT-Thread/rt-thread/pull/12013
- [23] Bug #11991 (ESP32-C3 v1.x rejects prebuilt images) — https://github.com/RT-Thread/rt-thread/issues/11991
- [24] PR #12058 (initial ArtInChip D213 support) — https://github.com/RT-Thread/rt-thread/pull/12058
- [25] PR #12054 (CAN framework refactor, under review) — https://github.com/RT-Thread/rt-thread/pull/12054
- [26] PR #12053 (CAN framework refactor first version, closed) — https://github.com/RT-Thread/rt-thread/pull/12053
- [27] Build repo PR #52 (QEMU virt config sync to 5.4.0) — https://github.com/RT-Thread/smart-build/pull/52
- [28] Build repo commit 21b1293c (defconfig sync) — https://github.com/RT-Thread/smart-build/commit/21b1293c
- [29] boa-z/rt-carplay (CarPlay receiver SDK) — https://github.com/boa-z/rt-carplay
- [30] boa-z/luban-lite (Luban-Lite snapshot) — https://github.com/boa-z/luban-lite
- [31] RT-Thread official news list — https://www.rt-thread.org/news.html
- [32] RT-Thread official news API — https://club.rt-thread.org/api/index/getList?type=rt_org
- [33] RT-Thread org repository list — https://github.com/orgs/RT-Thread/repositories
- [34] releases — https://github.com/RT-Thread/rt-thread/releases
- [35] PR list — https://github.com/RT-Thread/rt-thread/pulls
- [36] Issue list — https://github.com/RT-Thread/rt-thread/issues
- [37] Main repo commit history — https://github.com/RT-Thread/rt-thread/commits/master
- [38] Main branch version header — https://github.com/RT-Thread/rt-thread/blob/master/include/rtdef.h