Finished Phase 1: removed custody transfer, no admin record processing. - #110
Open
scburleigh wants to merge 181 commits into
Open
Finished Phase 1: removed custody transfer, no admin record processing.#110scburleigh wants to merge 181 commits into
scburleigh wants to merge 181 commits into
Conversation
The bounds check in both the SDR read (_sdrfetch) and write (_sdrput) primitives computed `to = from + length` before comparing against dsSize. That addition can wrap (from/length are uaddr/size_t), letting a crafted from/length pair slip past the `to > dsSize` guard and read or write outside the dataspace. Reorder to the overflow-safe form `length > dsSize || from > dsSize - length`, evaluated before any addition, and drop the now-unused `to` local. Ports the public ION-DTN _sdrfetch hardening and extends it to the matching _sdrput write path.
…C 9171 CTEB custodian_eid, CREB source_eid/report_to_eid, and the Bundle Sequence source EID item were encoded as CBOR text strings, diverging from the RFC 9171 structured [uri-code, SSP] array form required by Orange Book Draft K (CCSDS). Add serializeEidString() and acquireEidString() helpers in libbp.c that wrap the existing primary-block EID codec (serializeEid/acquireEid plus parseEidString/jotEid/readEid), inheriting correct 2-tuple and 3-tuple IPN handling. Replace the cbor_encode_text_string / cbor_decode_text_string calls in cteb.c, creb.c, and cbr.c with these helpers. Declare both in bp.h and bpP.h. Remove tests/pylib/cteb.py, dead scaffolding from a removed era with no callers in the current tree. Verified with custody-simple end-to-end test (2-node custody transfer with CCS round-trip): 2 passed, 0 failed.
Two follow-up fixes to the Orange Book wire-format work: CREB EID persistence: CrebScratchpad stored sourceEid and reportToEid as SDR Object handles, but creb_parse always set them to 0 after decoding, dropping the EIDs before creb_record could persist them to SDR. Change both fields to embedded char arrays (matching the CTEB pattern), so creb_parse can istrcpy the decoded EID directly into the scratchpad. Simplifies creb_release (no SDR strings to free) and creb_copy (plain sdr_write suffices for embedded arrays). seq-id-ref per-destination mode: per Orange Book CDDL, seq-id-ref is (uint .gt 0) / eid. When seqId == 0 (per-destination counter mode), cbr_encodeBundleSequence now serializes the destination EID as the CBOR item 1 instead of integer 0. cbr_decodeBundleSequence peeks at the CBOR major type of item 1: if it is an array it decodes a structured EID (setting seqId = 0) and returns it via new char **seqDestEid output parameter; otherwise it decodes an integer seqId as before. Update cbr.h declaration and all callers (cbr_handleCcs, cbr_handleCrs, bptrace.c). Verified with custody-simple end-to-end test in container: 2 passed, 0 failed.
…ession test Add m crebexpliciteid <0|1> to bpadmin so CREB blocks can be forced to include an explicit source EID (arrayLen=4), enabling ION-to-ION testing of the creb_parse EID persistence path that was fixed in the prior commit. Add tests/cbr-ct-orange-book/creb-eid-persist: a 3-node relay test that sets crebexpliciteid on the source node, sends a bundle requesting CRS signals for rcv+dlv, and verifies that Node 1 receives CRS back, Node 3 delivers the bundle, and no [?] CREB parse errors appear at any node.
Verifies the deletion CRS path in sendCompressedStatusRpt: when a bundle with the del-report flag is forwarded to a relay that has no route to the destination, the relay abandons with SrNoKnownRoute and sends a deletion CRS (status=3) back to the report-to EID. Node 1 loads cgr-hint.ionrc (contacts 2<->3) so its CGR computes path 1->2->3 and forwards the bundle to Node 2. Node 2 omits cgr-hint.ionrc, so its ipnfw immediately abandons with SrNoKnownRoute rather than holding the bundle for a bpclock cycle (~60 s) waiting for an outduct. Three checks: CRS Signals Recv increments at Node 1, "[i] CRS received: status: 3" in Node 1's ion.log, no "[?] CRS" parse errors at either node.
…ray test createBundle in libbpP.c copied most BpAncillaryData fields but silently dropped cbrSeqId, so every bundle got seqId=0 regardless of what the caller passed. This made creb_offer always encode seqId=0 in the CREB block, forcing all CRS aggregation into per-destination sequence counters even when a global seqId was requested. Add the missing cbrSeqId copy to createBundle. Add range-array count logging to cbr_handleCrs so decoded non-contiguous CRS signals are visible in ion.log. Add tests/cbr-ct-orange-book/seqid-gap: a 3-node regression test that sends 4 bundles sharing global seqId=1, drops one mid-stream to create a gap in the sequence numbers, and verifies that the aggregated forwarding CRS uses CBOR range-array encoding rather than a simple contiguous length.
Add opt-in timer-based custody retransmission to the Orange Book CBR implementation. By default retransmission is disabled (CBR_RETX_NONE); operators enable it via a new bpadmin command: m cbrretx timer <interval-seconds> <max-retransmissions> The new manageCbrRetx function in bpadmin.c parses the command and calls cbr_configureRetransmission. A second SDR transaction block is added to cbr_processTimeouts in cbr.c that walks the custodyBundles list once per bpclock second; any entry whose lastTransmit age exceeds retransmitIntervalSec (and whose retransmitCount is below the configured cap) is reforwarded via bpReforwardBundle and logged with "CBR: Timer-triggered retransmit". Add regression test custody-retransmit-timer: Node 1 sends one custody bundle to Node 2, which has no return path for the CCS. After the 5- second interval the timer walk fires and the test confirms the retransmit log entry appears within 12 seconds.
Add per-dimension custody acceptance policy to the Orange Book CBR-CT layer. Two independent SDR lists in CbrDb — one keyed by custodian EID (the node requesting custody transfer) and one by bundle source EID — act as whitelists. An empty list means accept from anyone (default, backward-compatible). When either list is non-empty, an incoming custody request whose EID is absent from that list is refused with a CCS refusal signal. New public API in cbr.c/cbr.h: cbr_addCustodyAccept, cbr_removeCustodyAccept, cbr_getCustodyAcceptList, and cbr_isCustodyAccepted. The policy check is applied in cteb_processOnAccept before cbr_acceptCustody. New bpadmin commands: a/d/l cbraccept custodian|source <eid>. bprc.pod updated to document cbraccept, cbrretx, and crebexpliciteid (the latter two were previously undocumented).
Two-phase regression test for the cbraccept whitelist policy:
Phase 1 (refusal): Node 1 starts with custodian whitelist {ipn:9.0}.
Node 2 sends a custody bundle; Node 1 refuses because ipn:2.0 is not
whitelisted and sends a CCS refusal.
Phase 2 (acceptance): ipn:2.0 is added to Node 1's custodian whitelist
via bpadmin at runtime. Node 2 sends another bundle; Node 1 accepts.
Covers the a/d/l cbraccept bpadmin commands, the SDR whitelist
persistence, and the cteb_processOnAccept policy enforcement path.
Replace UTF-8 em-dashes with ASCII -- in the cbraccept section. pod2man requires =encoding utf8 before any non-ASCII characters; using ASCII dashes avoids the issue consistently with all other pod files in the tree.
Add a custodyReqDests SDR list to CbrDb that holds destination EIDs eligible for automatic custody transfer promotion. When bpSend is called with NoCustodyRequested and Orange Book custody mode is active, the destination EID is checked against the list; if matched, custodySwitch is promoted to SourceCustodyRequired before CTEB construction, triggering full CBR-CT processing without requiring the application to set the flag. New bpadmin commands: a custodyreq <eid>, d custodyreq <eid>, l custodyreq. New API in cbr.c/cbr.h: cbr_addCustodyReq, cbr_removeCustodyReq, cbr_isCustodyRequired, cbr_getCustodyReqList. Add regression test custody-auto-request that verifies policy-active (custody accepted at Node 1, CCS returned to Node 2) and policy-removed (no custody accepted) phases, with no ION errors in either phase.
cbr_handleCrs now stores one ReceivedCrsRecord per status-code entry in a
per-node SDR ring buffer (CbrDb.crsHistory, default max 100). The sender
EID is captured from dlv.bundleSourceEid at the libbpP.c call site. When
the ring is full the oldest entry is evicted.
New bpadmin commands:
l crslog [<eid>] - list received CRS records newest-first, with optional
sender-EID filter
m crsmaxlog <n> - set ring-buffer limit (0 = unlimited)
cbr_getCrsHistoryList() exposes the SDR list to bpadmin using the same
getter pattern as cbr_getCustodyReqList(); cbr_setCrsHistoryMax() updates
the limit. The old cbr_listCrsHistory() (which used printText, unavailable
outside bpadmin) is removed.
bprc.pod updated with l crslog and m crsmaxlog documentation.
Regression test: tests/cbr-ct-orange-book/crs-history (8/8 pass).
Three changes complete the end-to-end path for custody-event CRS: creb_offer: now attaches a CREB block when BP_CT_REQUESTED is set, even when no traditional SRR flags are present. Previously CREB was skipped for custody-only bundles. mapSrrToCrebFlags: maps BP_CT_REQUESTED to CREB_REQUEST_CUSTODY_ACCEPT | CREB_REQUEST_CUSTODY_REFUSE. Bundles requesting custody automatically ask the custodian for custody-event CRS. cteb_processOnAccept: after each custody decision (acceptance or whitelist refusal), calls sendCustodyCrsIfRequested() which reads the CREB requestFlags and -- if the matching bit is set -- calls cbr_reportStatus with CBR_STATUS_CUSTODY_ACCEPTED (4) or CBR_STATUS_CUSTODY_REFUSED (5). The CRS goes to bundle->reportTo and is stored in the G2 CRS history log at the receiving node. creb_getRequestFlags() added to creb.c/creb.h to read the requestFlags byte from the CrebScratchpad. Regression test: tests/cbr-ct-orange-book/creb-custody-flags (6/6 pass).
…ecode Two bugs in cbr_handleCcs in cbr.c: 1. Loop variable shadowing — the signal-triggered retransmit inner loops for both the contiguous-range path and the range-array path reused the outer map-level loop variable 'i' as their own counter. Added uvast k and renamed both inner counters to k so the outer for-i-mapLen loop is not corrupted. The bug is latent with the current ION CCS encoder (mapLen is always 1) but would silently skip disposition entries in a conformant multi-disposition CCS from a third-party implementation. 2. Refused CCS CBOR decode failure — the disposition key was decoded with cbor_decode_integer (unsigned only), causing a hard CBOR error and ipnadminep crash when the disposition value is the CBOR negative integer -1 (refused). Replaced with cbor_decode_signed_int which handles both major type 0 (accepted +1) and major type 1 (refused -1). Regression test: tests/cbr-ct-orange-book/ccs-retransmit-range-array
Surface the cbr_listCustodyBundles and cbr_getCustodyStatus APIs (cbr.c)
through two new bpadmin commands:
l custodybundle - lists every bundle currently held in custody tracking
(seqId, seqNum, next-custodian EID, accept time, last-retransmit time,
retransmit count).
i custodybundle <sourceEid> <seqId> <seqNum> - reports the custody status
of a specific bundle: pending (in tracking, awaiting next-hop CCS) or
not-found (released, expired, or never tracked).
Both require Orange Book custody mode. Also document the previously
undocumented l crslog command. Man page bprc.pod updated.
Regression test tests/cbr-ct-orange-book/custody-bundle-query (2-node,
RETX_NONE plus cbraccept whitelist refusal) passes 6/6.
Let a node redirect all its Compressed Reporting Signals to a fixed collection EID instead of each bundle's primary reportTo field. New bpadmin commands m crebreportto <eid> / m crebreportto - set and clear a per-node override stored in CbrDb.crebDefaultReportToEid (cbrP.h); l crebreportto displays it. Accessors cbr_setCrebReportToEid and cbr_getCrebReportToEid added in cbr.c. Source side: when the override is set, creb_offer (creb.c) emits it as CREB element 4 (reportToEid), forcing array length 5 and including the source EID (element 3) that a length-5 array requires. Receiving side: new creb_getReportToEid (creb.c) extracts element 4 from a received CREB block. Both CRS-generating paths now honor it when building the CrebBlk for cbr_reportStatus - sendCompressedStatusRpt (libbpP.c) for delivery/receipt reports and sendCustodyCrsIfRequested (cteb.c) for custody-event reports - so the CRS is addressed to the override rather than bundle->reportTo. Requires m srmode compressed (or both). Man page bprc.pod updated. Regression test tests/cbr-ct-orange-book/crebreportto-redirect (3-node) passes 7/7: Node 1 overrides to ipn:3.0, the delivery CRS arrives at Node 3 and not at Node 1. The test runs a bpsink on Node 2 because a delivery status report only fires on actual delivery to an application - without a receiver the bundle is discarded and no CRS is generated.
Per Orange Book Section 3.2.9 a bundle sequence counter may wrap at a
16-, 32-, or 64-bit maximum, and matching a constrained peer's width is
required for End-to-End Gap Detection. Previously counterMaxValue was
hardcoded to CBR_COUNTER_MAX_64BIT in cbr_createSeqCounter with no way to
change it.
Add a node-wide default in the new CbrDb.counterMaxValue field (cbrP.h),
defaulted to 64-bit in cbr_initialize. cbr_createSeqCounter now seeds each
new counter from this default (falling back to 64-bit when the field is
zero). New accessors cbr_setCounterMaxValue / cbr_getCounterMaxValue in
cbr.c; the setter accepts only the three valid CBR_COUNTER_MAX_* values.
New bpadmin commands in bpadmin.c: m cbrcounterwidth { 16 | 32 | 64 } maps
the bit-width to the matching maximum and rejects other values;
l cbrcounterwidth shows the current width and maximum. The width applies
to counters created after the command; existing counters are unchanged.
Man page bprc.pod updated.
Regression test tests/cbr-ct-orange-book/cbr-counter-width (single node)
passes 5/5: startup 16-bit, runtime change to 32- and 64-bit, and
rejection of an out-of-range width.
The aggregation limits (crsAggregateLimit, ccsAggregateLimit, aggregateTimeoutSec) were read through the cached _cbrConstants() snapshot, which each process populates once at startup. A running daemon - notably bpclock, which runs the periodic timeout sweep - never saw a runtime m cbraggr, so the change only took effect after a restart. Add cbr_liveAggregateConfig in cbr.c, which sdr_reads the limits from the CbrDb inside the caller's transaction, and use it at the three decision points: cbr_processTimeouts (both timeout loops), cbr_reportStatus (CRS immediate-transmit check), and queueCcs (CCS immediate-transmit check). cbr_getConfig now also reads straight from the SDR so reported values are live. Add l cbraggr (bpadmin.c) to display the current CRS limit, CCS limit, and timeout; manageCbrAggr moved ahead of executeList so both dispatchers can reach it. Man page bprc.pod updated to note runtime effect and the l form. Regression test tests/cbr-ct-orange-book/cbr-aggr-runtime (2-node) passes 6/6: Node 2 starts with a 1-hour timeout so a generated delivery CRS is held pending; after m cbraggr lowers the timeout at runtime, bpclock flushes the pending CRS to Node 1. Verified no regressions in the crebreportto-redirect, custody-bundle-query, and cbr-counter-width tests.
The CBR/CT tutorial documented only m custodymode and m cbraggr and leaned on the cbrcustodytest developer utility for monitoring; it also was not referenced in mkdocs.yml, so it never appeared in the site navigation. Add a Command Reference section covering every Orange Book bpadmin option, grouped as mode/reporting (m custodymode, m srmode), aggregation (m/l cbraggr), retransmission (m cbrretx), CREB (m crebexpliciteid, m/l crebreportto, m/l cbrcounterwidth), custody-acceptance whitelist (a/d/l cbraccept), auto custody-request policy (a/d/l custodyreq), and monitoring/history (l/i custodybundle, m crsmaxlog, l crslog), each with syntax, examples, and defaults, plus a startup-vs-runtime note. Refresh the monitoring, retransmission, and testing-chain sections to use the supported bpadmin commands (l/i custodybundle, l crslog, m cbrretx) instead of cbrcustodytest, which is now described as a developer test tool. Add the page to mkdocs.yml under Configuration as "CBR/CT (Orange Book) Tutorial" so it is reachable from the site navigation.
Extract atomic tier computation, CFLAGS configuration, and RTEMS QEMU verification into standalone scripts. This reduces workflow file complexity and makes the logic testable and reusable across workflows. - compute-atomic-tiers.sh: Dynamic test tier assignment - compute-tier-cflags.sh: Tier-specific compiler flags - rtems-verify-qemu-output.sh: QEMU output validation 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
Replace timestamp-based cache keys with content hashes for more efficient caching. RTEMS build artifacts and atomic tier builds now use file content hashing to determine cache validity, improving cache hit rates and reducing unnecessary rebuilds. - RTEMS: Cache based on source file hashes - Atomic tiers: Content-hash keys with workflow script integration - Enable cache reuse across workflow runs when source unchanged 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
Enhance test summary script to handle runner names with dashes, preserve test logs, and generate compact JSON for GitHub Actions. Externalize PR status logic to the summary script for better separation of concerns. - Robust runner name parsing - Comprehensive test log preservation - Compact JSON output for artifacts - Bundled artifact handling 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
Introduce build-runner-matrix.sh for dynamic runner selection. Improve multi-platform test summary generation. - Dynamic runner matrix based on platform requirements - Enhanced test summary with multi-platform support 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
Align ARC and Solaris workflows with project-wide naming conventions: test_batch replaces test_group, num_runners replaces runner_count. Update generate-test-matrix action for consistency. 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
Configure PR test workflow to use direct timestamp references and support automatic execution. Streamlines CI/CD process for pull request validation. 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
This was only used for debugging purposes. 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
Improve EXTRA_CONFIGURE_FLAGS handling in compile-and-buildcheck script with proper word splitting 🚀 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
The LTP duct addresses and the LTP engine number appear in several CLA-setup calls; extract them into TEST_LTP_DUCT and LTP_ENGINE_STR so each value has a single definition (and the engine number's required match with ION_NODE_NBR is documented).
ShellCheck warns about these ln statements because they don't have an explicit destination [1]. Seems like POSIX ln requires a target operand [2]. Also change shebang line to #!/bin/sh to help in our goal of removing Bash as a dependency. [1]: https://www.shellcheck.net/wiki/SC2226 [2]: https://pubs.opengroup.org/onlinepubs/9799919799/utilities/ln.html
The srclinks script invokes this srcremove script to remove all symbolic links. While it replaces the C source files and header files, it does not replace rtems_waf. So exclude rtems_waf from deletion, otherwise we'd have to manually add it back ourselves. Shebang line is also changed to #!/bin/sh in line with the goal to remove Bash as a dependency.
Wires tcpcli/tcpclo into the minimal RTEMS port and runs a TCP loopback test alongside the existing UDP/LTP one.
PSM small-pool blocks are aligned to WORD_SIZE (4 bytes on 32-bit), so 8-byte-aligned types (long long, time_t) embedded in PSM-allocated structs traps on RTEMS 6 sparc/leon3. Introduce PSM_BLK_ALIGN -- the larger of WORD_SIZE and 8 -- as the small-block grain/overhead so user data is always suitably aligned for any basic C type. Identical to the old behaviour on LP64; lets the RTEMS port keep LONG_LONG_OKAY=1 on every BSP.
The SIGTERM -> grace-wait -> SIGKILL escalation is used to reap snooze-based daemons at shutdown (needed on RTEMS, where the SIGTERM handler's sm_TaskVar flag is not updated from signal context) is copy-pasted across many stop paths. Extract it into one ICI helper, sm_TaskKillWait(taskId, text, note), and use it in bpStop and the scheme/plan/induct/outduct waiters (libbpP.c), ltpStop / waitForSpan / waitForSeat (libltpP.c), bsspStop and its span/seat waiters (libbsspP.c), rfx_stop (rfx.c), and cfdpStop.
For a root-level dest ("/foo.dat") getQualifiedFileName's last-separator
scan finds none (it stops short of the leading '/'), so the whole path
was mkdir'd as a directory -> EISDIR on open. Return early when there's
no interior separator (parent already exists). Also route cfdpStop
through sm_TaskKillWait().
Commit 45e46e9 (Resolved reported concurrency issues in udplsi.c / udplso.c: add mutex usage, join threads, remove close() race, 2025-03-19) made rtp a static variable which breaks the alignment of the declaration names. Run clang-format on these lines to align them again.
We perform a platform test to guess whether the iphdr, udphdr, udpiphdr structures are available. This isn't robust (FreeBSD and libbsd has the ip structure in netinet/ip.h) and it goes against our goal to test for features instead of guessing based on platforms. It's also unnecessary. We know, a priori, the sizes of these UDP, IPv4, and IPv6 headers. Commit 4f05601 (prototype LTP with dual stack ipv6 and ipv4 support with regression test, 2025-10-15) added support for IPv6 to udplso, but did not update this header size logic to handle IPv6. Remove the platform checks, remove the macros using the structures, and just use named constants with the known header sizes. Also rework udplso to use different sizes based on the address family.
pseudoshell passes task argument string pointers through FUNCPTR, and on 64-bit RTEMS targets (riscv rv64, aarch64) an int parameter would truncate them. Use saddr (pointer-sized) instead.
Single build script that fetches/builds the RTEMS kernel and rtems-libbsd and compiles the ION BPv7 minimal port for the supported BSPs (sparc/leon3, aarch64, arm, riscv, powerpc), then runs the loopback tests under QEMU.
With configuration "format.quote-style = 'single'" to minimize diffs since that's the predominant style in this file.
The Waf script assumes SPACE_ORDER=3 which is incorrect for 32-bit systems. Set SPACE_ORDER based on the architecture.
Really don't like the idea of disabling warnings to avoid dealing with parseEidString() incorrectly assuming LONG_WIDTH == 64 and cbr.h incorrectly assuming uvast is 64 bits wide. That code should just be fixed. Disabling these warnings can also mask other issues. But given that 32-bit architectures aren't official targets for ION, we decided it was acceptable to disable these and file an issue to address the wrong assumptions at a later time.
Tasks use the RTEMS_NO_FLOATING_POINT attribute by default. A RTEMS_NO_FLOATING_POINT task attempting to access the FPU on LEON3 results in an exception.
The deep admin call chain in startDTN() allocates large local variables on the stack. The default CONFIGURE_MINIMUM_TASK_STACK_SIZE is insufficient, causing a silent overflow that corrupts adjacent libio structures and results in ENXIO filesystem errors. 128 KiB seems to be enough to address this.
bpcp and bpcpd are removed from cfdpsources because they're user-facing scp-like utilities that depend on CFDP proxy/remote operations (CfdpProxyTask, cfdp_rput, cfdp_get, cfdp_rls) which require -UNO_PROXY. We keep NO_PROXY defined in this minimal port, so they're omitted; cfdpadmin/cfdpclock/bputa are all the ionrtems.c smoke test needs. Commit acdaa6c (Updating Makefile.am and configure.ac to reflect the change to INB to build cfdp and ams., 2018-03-05) made CFDP unconditionally built in ION's regular build system. To avoid discrepancies, the RTEMS port will also build CFDP unconditionally. ION_OPEN_SOURCE is removed from srclinks since it has no effect. (ION_OPEN_SOURCE and NASA_PROTECTED_FLIGHT_CODE were supposed to be scrubbed in 4.1.3s).
AMS and CFDP used to be built conditionally. The IS_NASA_B variable at configuration time set the ION_NASA_B Automake conditional and defined the NASA_PROTECTED_FLIGHT_CODE macro to conditionally guard AMS and CFDP code. AMS and CFDP were made unconditional in commit acdaa6c (Updating Makefile.am and configure.ac to reflect the change to INB to build cfdp and ams., 2018-03-05), and NASA_PROTECTED_FLIGHT_CODE was mostly removed by 4.1.3s. ionexit and ionrestart still used this macro since the RTEMS port used to build CFDP conditionally. Now that CFDP is unconditional on RTEMS, we can finally clean up the last usages of NASA_PROTECTED_FLIGHT_CODE.
AMS is resource intensive. We have to bump the max number of POSIX Threads for the init task (must be greater than the platform_sm MAX_POSIX_TASKS set in the Waf script). SDR size and working memory size are also bumped to seemingly sufficient levels. Since we don't have name resolution, loadTestMib()'s usage of getNameOfHost() won't work properly since the result (e.g., localhost:2357) won't resolve. So we need to set the hostname to be an IP address.
Remove "Hello World" placeholder and list the BSPs we support.
The RTEMS QEMU loopback test now exercises TCP/TCPCL, CFDP, and AMS in addition to the baseline UDP/LTP loopback. Extend the externalized verifier rtems-verify-qemu-output.sh to assert each added scenario ran and delivered, so the script-based verification matches the expanded coverage of the test in ionrtems.c. The RTEMS multi-test runs UDP/LTP, TCP/TCPCL, CFDP, and AMS in sequence. AMS is last and needs a short registration retry window, so it does not deliver until ~40s in; the previous 30s QEMU timeout killed the run before AMS completed. Extend the timeout to 60s so all scenarios finish. Also make the bundle xmt/rcv statistics checks in rtems-verify-qemu-output.sh honor the --qemu-output argument instead of a hardcoded rtems-test-output/qemu-output.txt path, so the verifier works when invoked against an output file elsewhere.
gh-pages doc: restructure into Verified / Community-contributed / Known broken sections; list the full set of modules now exercised on the verified target (BP, LTP, TCPCL, CFDP, BSSP, DGR, AMS); add a footprint note about the PSM small-block alignment widening on 32-bit RTEMS targets. Streamlined the RTEMS 6.1 port documentation by: - Removed obsolete RTEMS 5.1 and 4.10.2 archive sections (130+ lines) - Fixed incorrect dates (was "11/7/2025", now current) - Corrected version reference (4.1.5 -> 4.1.4+) - Reduced from 515 to 374 lines while preserving all essential info - Improved readability with better organization - Kept all technical details users need for adaptation The README now focuses exclusively on the current RTEMS 6.1 ARM64 port without historical clutter. All build instructions, configuration details, and implementation notes remain intact.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.