Research window: 2026-11-10 08:00 ~ 2026-11-11 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 (35 repositories checked in push order), Google News Chinese and English queries (via proxy, including three-day window and bare query comparison), HN Algolia, Electronic Engineering Album RT-Thread column, club.rt-thread.org forum


Issue Index

  • I. Kernel and Driver Merges
    • RISC-V APLIC: No More MSI Resend After Level Interrupt Withdrawal (11-10)
  • II. Security Hardening Batch: Eleven Boundary Checks in One Day
    • Six 2024 CVEs Closed Out in a Batch (11-10)
    • Loaders and Kernel Objects: libdl Relocation, LWP IPC Owner Validation (11-10)
    • Memory and Networking: Memory Pool Assertion False Positive, DHCP Lease Expiry Reclamation (11-10)
  • III. Issues and Fix Closure
    • ptyfs Handle Leak: Fix Issued the Same Day as Report (11-10)
    • legacy USB host: One Fix Closed by Maintenance Policy (11-10)
    • vDSO Clock Constant: Issue Accepted, Fix Rejected (11-10)
  • IV. Ecosystem and Vendor Adaptation
    • ESP32-S3 Port Half-Landed: Build Script Committed, BSP Not in Repo (11-10)
    • packages and Toolchain Repos Had No Pushes This Period (11-11)
  • V. Official and Community Activities
    • Official News API Has No New Items for the Third Consecutive Period (11-11)
    • Google News Three-Day Window Zero Hits: Channel Normal, Genuinely No New Articles (11-11)
  • VI. Summary and Trend Observations
    • Security Review Shifts from Individual Cases to Batches
    • Six CVEs and a Two-and-a-Half-Year Time Gap
    • Pending Queue: 149 Proposals, 365 Issues
    • Fix Closed by Policy: The Boundary of legacy Code

I. Kernel and Driver Merges

1.1 RISC-V APLIC: No More MSI Resend After Level Interrupt Withdrawal (11-10)

Date: 2026-11-10 16:58 (merge time) Source: Proposal #11966

One file, 14 lines added, 1 line deleted, modifying libcpu/risc-v/common/pic/pic-riscv-aplic.c, authored by zijunteng238-dot, submitted November 1, merged this period. The previous report named this item in the pending queue; it landed today.

The symptom was running bsp/qemu-virt64-riscv on QEMU 8.2.2: RV64, quad-core SMP, AIA enabled (APLIC/IMSIC), a counter program printing every 1000 milliseconds output only “RISC-V test: 0” and then stalled. The root cause lies in the EOI path for level interrupts—the driver unconditionally writes setipnum_le, but the device had actually withdrawn the interrupt; older QEMU still resent an MSI based on this write, creating a persistent spurious interrupt that suppressed the system timer, and the delay thread could never return. The fix reads the rectified state of that input in in_clrip (already converted per interrupt polarity) at EOI, and only writes setipnum_le if the device is still asserting, preserving the level interrupt’s retrigger capability while giving no resend opportunity to a withdrawn source.

The same proposal changes qemu-virt64-riscv’s main.c from printing once to a counting loop from 0 to 9, delaying one second after each print, to observe thread recovery after the fix. Verification covered three configurations: AIA enabled quad-core with networking/NVMe/virtio-console/PCI serial disabled, AIA disabled quad-core, and AIA enabled quad-core with default device combination—all three produced the full ten lines; real APLIC hardware, active-low peripherals, and other QEMU versions were not covered.

QEMU upstream has already fixed this emulation defect itself; this RT-Thread change is a compatibility patch for older emulators—it fixes not a kernel logic error, but a mismatch that only appears on virtual platforms, where “the driver writes per spec, the emulator interprets per defect.”

II. Security Hardening Batch: Eleven Boundary Checks in One Day

2.1 Six 2024 CVEs Closed Out Together (11-10)

Date: 2026-11-10 01:04 ~ 06:45 (proposal time) Source: Proposal list

Within a single day on November 10, the same reporter (cms19859230182-lang) submitted eleven boundary-check patches, six of which map directly to CVEs disclosed in 2024. The numbers are adjacent and the fixes are structurally identical—all follow the pattern “check the length first, then copy”:

  • Proposal #12017: FinSH builds directory paths with unbounded rt_sprintf / rt_strcpy / strcat. Auto-completion now reallocates according to DFS_PATH_MAX and caps each segment copy, and the length check is changed back from > to >= (restoring the off-by-one for the NUL). Corresponds to CVE-2024-25390, issue #8286.
  • Proposal #12020: The len parameter of rt_wlan_send_to_thread() is a signed int; a negative value causes sizeof + len and memcpy to be treated as a huge size_t. The fix returns -RT_EINVAL directly when len < 0. Corresponds to CVE-2024-25388, issue #8285.
  • Proposal #12022: rt_random is a predictable LCG with a weak default seed. The first read now mixes in rt_tick_get() and an optional hardware random source, and comments and Kconfig explicitly state it is not a CSPRNG. Corresponds to CVE-2024-25389, issue #8283.
  • Proposal #12019: var_export fills ve_exporter_tab[2048] with no capacity limit; with assertions disabled in production builds it writes straight past the end. The fix stops when full, clamps the count at runtime, and skips the bubble sort when there are zero elements. Corresponds to CVE-2024-25392, issue #8290.
  • Proposal #12018: rt-link does not validate the final frame length before framing, so an oversized payload writes past the static sendbuffer. The fix checks data_len and the frame length before memset/memcpy. Corresponds to CVE-2024-25395, issue #8289.
  • Proposal #12021: YMODEM’s _rym_send_begin writes to the protocol buffer with unbounded rt_sprintf, and upload/download use rt_strncpy(..., DFS_PATH_MAX), which may not append a NUL. The fix uses rt_snprintf with a cap of sizeof-1 and explicitly zero-fills. Corresponds to CVE-2024-25394, issue #8291.

What these six share is “using assertions as boundary checks”: RT_ASSERT is often disabled in production configurations, and many places previously treated assertions as the last line of defense—which in practice meant no check at all. This batch converts the limits into runtime checks, leaving assertions only as a debugging aid.

2.2 Loader and Kernel Objects: libdl Relocation, LWP IPC Ownership Validation (11-10)

Date: 2026-11-10 01:40 and 01:36 (proposal time) Source: Proposal #12016, Proposal #12015

These are the two most technically significant items in this batch, and both touch trust boundaries.

#12016 changes five files, with 205 lines added and 37 removed. When computing the relocation write target, dlmodule uses mem_space + r_offset - vstart_addr without validating the range; a crafted .mo can make r_offset underflow or overflow and write directly into kernel memory during relocation. The fix adds dlmodule_check_rel_target(), which uses overflow-safe arithmetic to confirm the target range falls within module memory, and wires it into the ARM, RISC-V, and x86 relocators. It also adds checks for ELF section header indices (sh_link, e_shstrndx), symbol table indices (ELF*_R_SYM), and ET_REL section bounds, and makes dlelf.c abort loading and free section_load_addr on failure. Corresponds to defect #11898 from October 18.

#12015 changes three files, with 284 lines added and 44 removed. RT-Smart’s LWP IPC system calls previously passed raw user-supplied object pointers straight into rt_sem_* / mutex / event / mailbox / mq / timer, so a fake object could cause arbitrary kernel reads and writes. The fix adds lwp_user_object_get() and lwp_user_object_put(): before dereferencing, reject user-space addresses, require the handle to exist in the caller’s user-object AVL tree, compare rt_object_get_type(), and hold a reference for the entire system call (including all error paths); lwp_ref_count now covers both the owner and in-flight references, and a kernel object is destroyed only after the last reference disappears. Corresponds to defect #11960 from October 31.

Neither item includes a constructed attack sample; validation is limited to static audit and compilation, which also means they will not land in a release in the near term.

2.3 Memory and Networking: Memory Pool Assertion False Positive, DHCP Lease Expiry Reclamation (11-10)

Date: 2026-11-10 01:04 and 06:45 (proposal time) Source: Proposal #12014, Proposal #12024

#12014 fixes a false alarm, but one that causes a real system stall: when rt_smem_free() frees “the last block, whose next points to the heap_end sentinel,” it incorrectly triggers the MEM_POOL assertion. The fix explicitly accepts the mem->next == heap_end offset while keeping the original validation for all other cases; the author explicitly rejects switching to a mem->prev check on the grounds that prev points to the previous block (0 for the first block), which would loosen the validation. An accompanying core.mem test case simulates the sentinel scenario with only the in-use bit set, verifying full-block coalescing. Corresponds to old defect #10632 from 2025.

#12024 fixes the DHCP server’s accounting: lease_end is written but never read, so leases never expire and addresses are never reclaimed. It also turns out that DHCP_DEFAULT_LIVE_TIME (0x80510100), after PP_HTONL(), is about sixty-eight years on the wire, completely at odds with the “reclaim by day” semantics, so it is changed to 86400 seconds as well. The change makes leases absolute expiry times, lazily reclaims expired nodes before allocation, and renews uniformly across the hit, new, and ACK paths. Validation consists only of syntax checks against lwIP 2.0.3 / 2.1.2 / 2.2.1.

This is a textbook case of “a configuration constant pointing the wrong way”: the macro name says lifetime, but the value treats a nanosecond-scale number as seconds—and it only went unnoticed because nobody actually waited for it to expire.

III. Issue and Fix Closure Loop

3.1 ptyfs handle leak: fix filed the same day as the report (11-10)

Date: 2026-11-10 14:04 (issue) / 11-11 02:14 (PR) Source: Bug #12023, PR #12029

ptyfs_unregister_pts() removes the pts file from the root list and reclaims the unit number, but never frees the struct ptyfs_file allocated by ptyfs_register_pts(). Once the unit number is reused, repeatedly calling open() / close() on /dev/pts/* leaks memory indefinitely, and an ordinary user can trigger it. Reporter XueDugu included a loop-based reproduction.

Twelve hours later, maintainer BernardXiong submitted #12029 (three files, 11 insertions, 2 deletions), which frees the pts file on the unregister path. It is currently awaiting review. This is the only item in this issue where the report and the fix closed within the same window.

3.2 legacy USB host: a fix closed by maintenance policy (11-10)

Date: 2026-11-10 18:13 (PR) / 11-11 00:50 (issue) Source: Bug #12026, PR #12027

The bug concerns the legacy USB host stack (components/legacy/usb/usbhost/): the device descriptor’s bLength is used as a read length without validation and then written into a static device array, allowing a malicious USB peripheral to cause an out-of-bounds write and control-flow hijacking. The reporter reproduced it end-to-end on qemu-virt64-aarch64 and noted that the main repo master (self-reported v5.4.0-dev, checked at commit 5d1fb66d) was still unfixed when re-read on November 11.

The maintainer then submitted #12027 (one file, 8 insertions) to add length validation, but immediately closed it with the comment: “This is legacy code and will no longer be continuously fixed or maintained.” The fix direction was never in dispute; the closure was based on maintenance scope—the known attack surface on this path is now left for users to assess themselves.

3.3 vDSO clock constants: issue valid, fix rejected (11-10)

Date: 2026-11-10 20:39 (issue) Source: Bug #12025, PR #12028

When compiling RT-Smart’s vDSO user code with a newlib toolchain, neither CLOCK_MONOTONIC nor the GNU clock IDs are defined: newlib only exposes the former under _POSIX_MONOTONIC_CLOCK and the latter under _GNU_SOURCE, and the vDSO’s freestanding build configuration sets neither. PR #12028 (one file, 22 insertions) chose to add local definitions, but the maintainer judged it “an incorrect fix that should be closed” and closed it the same day.

The divergence between the two approaches to the same symptom is clear: one adds macros in the compilation unit so the build passes; the other aligns the freestanding build configuration with the POSIX feature macros. The latter was not written into the PR, and the bug itself remains open.

IV. Ecosystem and Vendor Porting

4.1 ESP32-S3 port lands half-done: build script merged, BSP not in repo (11-10)

Date: 2026-11-10 11:53 (merge time) Source: PR #11844, bsp/ESP directory

A merge that is easy to misread. PR #11844 is titled “Fix the include path of xtensa_context.h,” but its body describes adding two directories, bsp/ESP/ESP32_S3 and libcpu/xtensa/esp32s3: ESP32-S3-DevKitM-1, Xtensa LX7, UART0 on GPIO43/44, 8MB DIO flash, application at 0x10000, and CI adding xtensa-esp32s3-elf GCC 12.2.0.

What actually merged was a single file: the SConscript under libcpu/xtensa/esp32s3/, 25 lines. Checking the main repo’s bsp/ESP directory, it currently contains only ESP32_C3, ESP32_C6, libraries, and tools—no ESP32_S3; and libcpu/xtensa/esp32s3 contains only this one script, with no C source files. The script itself resolves the ESP-IDF package path via bsp/ESP/ESP32_S3, which does not currently exist.

In other words, this is half a port: the CPU port’s build scaffolding entered the main repo first, while the BSP, tick handling, sdkconfig, and board-level validation are all absent. The PR was created on September 28 and merged on November 10; over those six weeks, the rest of what the body described did not land alongside it. The practical implication for users: there is currently no compilable ESP32-S3 BSP, and the path referenced by the script does not resolve for now.

4.2 No pushes to packages or toolchain repos this issue (11-11)

Date: 2026-11-11 (check) Source: RT-Thread org repositories

Sorting the 35 organization repositories by push time, only the main repository had a push within the window (11-10 16:58). The last push to packages was November 9 at 10:27 (the WCH SDK sync covered in the previous issue); env is stuck at 11-06, sdk at 11-05, and rtthread-nano, smart-build, pkgs-test, toolchains-ci, etc. all show no activity. The package index side saw zero changes today, unlike the paired-merge pace of the previous issue.

V. Official and Community Activities

5.1 Official news feed sees no additions for the third consecutive issue (11-11)

Date: 2026-11-11 (check) Source: Official news list

The feed still returns the same five articles from November 6: elexcon full-stack capabilities, the Vinaro industrial PLC salon, Southeast Asian university industry-education integration, the Lingang automotive software ecosystem salon, and the CIIF AI car Workshop, with the latest timestamp at 11-06 15:54. All of these were already covered in the 11-09 issue (the 72-hour window covering the weekend), and neither 11-10 nor this issue has additions—official external communications have entered a gap between events.

5.2 Google News three-day window: zero hits—channel healthy, genuinely no new articles (11-11)

Date: 2026-11-11 (check) Source: Google News

All three time-filtered queries (RT-Thread when:3d, 睿擎派 when:7d, RT-Thread 嵌入式 when:7d) returned zero hits. To distinguish “channel failure” from “genuinely no news,” two bare queries were run as controls: RT-Thread without a time filter returned 42 results (the latest being the 11-03 Wuzhen Summit article, long since covered), and RISC-V when:1d returned 8 news items from the day—so the channel itself is healthy, and RT-Thread genuinely had no new articles indexed during these three days.

All seven HN Algolia candidates were literal “thread” false matches (threaded UI, free-threaded Python, HN discussion-thread recommendation tools) unrelated to RT-Thread, and were all discarded.

VI. Summary and Trend Observations

6.1 Security review shifts from individual cases to batches

Date: 2026-11-10 Source: Pull request list

Of the 14 new pull requests in the window, 11 came from the same reporter and were submitted in a continuous burst over more than five hours in the early morning of November 10. The subjects are not scattered findings but a checklist swept module by module: FinSH, WLAN, random numbers, var_export, rt-link, YMODEM, libdl, LWP, memory pool, DHCP, ESP32-C3. Each includes three sections — “motivation / changes / verification” — with verification uniformly described as static checks and compilation, and an honest note that “no PoC was constructed.”

This shape differs from sporadic security issues: it looks more like the planned wrap-up of an audit, ranking years of accumulated low-severity items by exploitability and closing them one by one. For the project, the cost is a review queue that grew by nine entries in a single day, all requiring manual review.

6.2 Six CVEs and a two-and-a-half-year lag

Date: 2026-11-10 Source: Corresponding issues such as CVE-2024-25388

The original issue numbers corresponding to the six CVEs fall between #8283 and #8291, part of the 2024 report batch, and are only now being closed roughly two and a half years later. This need not be read as negligence: review resources for RTOS projects concentrate on chip adaptation and releases, and without someone sweeping back through the list, items like these — which “require constructing malformed input to trigger” — would remain at the bottom of the queue indefinitely.

What is truly worth noting is the consistency of the fixes — all six patch the same hole: treating assertions as bounds checks. Everyone in embedded projects knows assertions are compiled out in production builds, yet code still uses them as checks at scale.

6.3 Review queue: 149 pull requests, 365 issues

Date: 2026-11-11 (verification) Source: Pull request list, Issue list

Compared with 140 pull requests and 362 issues in the previous period, pull requests rose by nine and issues by three. The breakdown is clear: 2 merged, 14 new pull requests, and 2 closed after submission during the window — a net increase of nine; on the issue side, four were added (ptyfs, USB host, vDSO, PLIC ordering), one of which follows up on the PLIC issue already covered last period.

Several items called out last period remain unmerged: tty ioctl user-pointer validation (#11937, affecting all RT-Smart boards since v5.1.0, the highest risk), ivshmem window validation (#11993), RISC-V FPU context (#12006), and the CAN driver refactor (#11779). Today’s new #12015 and #12016 happen to cover the two kernel-level risks in #11960 and #11898, but both are still under review.

6.4 Fixes closed by policy: the boundary of legacy code

Date: 2026-11-10 Source: Pull request #12027

Two pull requests this period were submitted by maintainers themselves and closed the same day, and in neither case was the reason that the fix was technically invalid: one said the fix went in the wrong direction (a vDSO macro addition), the other that the code had entered an unmaintained state (legacy USB host). The latter deserves more attention — the defect description explicitly states that “a malicious USB peripheral can cause out-of-bounds writes and control-flow hijacking,” which is a physically reachable attack surface.

An RTOS that has run for twenty years inevitably carries legacy components, and declaring them unmaintained is a reasonable resource decision; but for products still using the old RT_USING_USB_HOST stack, that declaration effectively transfers security responsibility to the integrator. If the project documentation could list such “unmaintained but still in-tree” components as a checklist, it would be more useful than having them scattered across individual pull request comments.

Appendix: Source Verification Table

Source Verification Result
RT-Thread official news API Five items, latest 11-06 15:54, zero new items for the third consecutive issue
Official news page List is client-rendered; API is authoritative and matches the API
GitHub main repo commits Two within the window (APLIC fix, esp32s3 build script), checked by committer time
GitHub PRs and Issues Two merged, 14 new proposals, 4 new issues; 149 proposals and 365 issues pending review
RT-Thread org repositories 35 repositories checked sorted by push time; only the main repo had pushes within the window
packages repository No pushes; last at 11-09 10:27 (previous issue’s WCH SDK sync)
releases and tags No additions; latest remains v5.3.0 (09-10)
Google News Chinese Zero hits in the three-day window; bare query returned 42 results, confirming the channel is functioning
Google News English Zero valid hits
HN Algolia All seven candidates were literal “thread” false matches and were excluded
EET China RT-Thread column Unreachable this issue (connection failed); could not be verified
club.rt-thread.org forum Returned WAF block page (HTTP 468); could not be verified
bsp/ESP directory check Only ESP32_C3 and ESP32_C6; no ESP32_S3

Full Source List