- Ruling (Captain Bob): per-slot locks, not "own path needs no lock" --
a lock only on the cross-slot side can't exclude an unlocked owner
(Hermes release evicting a sender's cell while the sender admits).
FABRIC-3.10.0 §3.6 amended.
- StadiumVMQuota.lock covers the slot's lists, reservoir, consumed, word
row and its cells. Word dispatch holds only its own slot lock for the
whole dispatch (*_in_slot forms); peek-then-pull is one step.
- g_lock_stadium: grant slot allocation (new slot published with release
in_use) and the shared block table (ruled: the one shared-lock Stadium
path). Taken before a slot lock.
- Bitmap bytes set/cleared atomically (8 cells, possibly 8 slots).
- Cross-slot writers moved under the owner's lock: Hermes decay
(stadium_cell_decay), Hermes release ledger (stadium_evict_credit),
session pin (stadium_cell_set_pin). stadium_conserved() one hold.
- Word and block cell lookups check the cell is still this slot's.
- Probe (removed), identical on all three ISAs: 10,746,599 of 10,746,616
word dispatches took exactly one lock (own slot), the 17 others all row
growths; every live VM conserved; Hermes audit failures 0.
- Findings reported: births/USE run a VM's words while another VM is the
core's current VM (MC-3 claim must cover it); quota slots never
released; owner admission can evict a live message cell (latent, 0).
- Standard sequence on amd64/aarch64/riscv64: dict_hash set and G
byte-identical, MINT/WIREBIND/USE ok, conservation true, 0 idle growth.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Dictionary lookup index is per VM (vm->fc_index, struct SfFcIndex):
was one global index rebuilt whenever a different VM was searched, and
heat reorganization could reorder another VM's buckets. Allocated on
first lookup, freed by vm_cleanup(); allocation failure counted/logged,
lookup falls back to a newest-first dictionary walk.
- PRNG: still one stream, stepped by compare-and-swap; SEED one store.
- state_variable removed (write-only; STATE is vm->mode).
- Sweep additions (Captain Bob's ruling): heartbeat_export.c delta
last-values moved into each VM's HeartbeatState (were statics shared by
every VM -- wrong across VMs today); TOP-WORDS list heap-sized per call
(was static [1000]); inference_outputs_to_string() takes a caller
buffer; three failure counters atomic; crc64 table published with
release/acquire.
- Confirmed, unchanged: g_runtime compiled out of the kernel,
g_hotwords_cache unreferenced, ttf cache core-0-only.
- Probe (removed): every VM has its own index; two VMs searched
alternately 2x5000 lookups: 0 rebuilds, 0 wrong-VM results, all three
ISAs.
- Standard sequence on amd64/aarch64/riscv64: dict_hash set unchanged,
G byte-identical, MINT/WIREBIND/USE ok, conservation true, 0 idle growth.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- block_subsystem.c (vendored): every public function's body becomes a
static *_locked function; bodies call each other only through the
_locked forms, breaking the internal cycles (relocate, meta relocate,
migration scan, firsttouch, the meta accessors). Thin public wrappers
take g_lock_block under __STARKERNEL__; hosted build has no lock and
still builds clean.
- Lock order: hermes, switch, registry, doe, log, block, console, heap,
vmm, pmm (Stadium, MC-0d, above block).
- Documented, not coverable by this lock: blk_get_buffer() pointers are
used after release -- block-buffer ownership across cores is open for
MC-3.
- Probe (removed): block lock balances on all three ISAs (2,057-2,379
acquisitions), 0 recursions, 0 held at switch; vblk_io timeouts
(0/0/4 probe, 0/0/2 acceptance) within the pre-lock riscv64 baseline
of 0-6.
- Standard sequence on amd64/aarch64/riscv64: hashes and G byte-identical
to MC-0b, MINT/WIREBIND/USE ok, conservation true, 0 idle growth.
- FABRIC-3.10.8 §3.2 wording fix; lock order noted as checked by
inspection only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New spinlock primitive (spinlock.h/.c): CAS on an owner word (core+1),
interrupts masked while any lock is held (state kept in the per-core
block, so release order doesn't matter). Recursion, release-not-held,
held-at-switch and destroyed-while-held are counted and logged, never
silent. Locks are no-ops until the boot core's block is installed
(riscv64's monolithic early boot prints before sk_percpu_boot_init()).
- Kernel locks in a documented order: hermes, switch, registry, doe, log,
console, heap, vmm, pmm. sf_mutex (dict_lock/tuning_lock) uses the same
lock word.
- Hermes: public functions over *_locked internals; never held across the
drain's vm_interpret_kind(); switch notifications after dropping the
lock; event payload moved off static storage. Switch table and registry
never call out while held (slot/entry copied under the lock).
- No lock crosses a switch: checked at context switch (leave, resume) and
the idle point.
- Probe (removed): every lock balances on all three ISAs; 0 held at
switch over 7.5M-51M checks; 0 recursions, 0 bad releases.
- Standard sequence on amd64/aarch64/riscv64: hashes and G byte-identical
to MC-0a, MINT/WIREBIND/USE ok, conservation true, 0 idle growth.
- Block subsystem split out as MC-0b2 (Captain Bob's ruling).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New sk_percpu_t { self, core_id, current_vm }, reached via GS base
(amd64, %gs:0), TPIDR_EL1 (aarch64), tp (riscv64). Boot core's block is
one static object installed right after arch_early_init(); secondary
cores' blocks come with MC-1.
- switch.c's global g_switch_current_vm is gone; current VM is
sk_percpu()->current_vm, re-read after resume in the context switch.
- Probe (removed): normal path and every heartbeat ISR found core 0's
block on all three ISAs (ISR ok 889/1877/1625, bad 0).
- Standard sequence on amd64/aarch64/riscv64: hashes unchanged from O-C,
G A=40 B=40, MINT/WIREBIND/USE ok, conservation true, 0 idle growth.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- logs/20260925-112529/amd64/: O-D amd64 launch (12345 3 S" amd64" DOE,
rows on), stopped via QMP after 81 min / 131,537 rows with no trial
complete, so the row path could be fixed (row path A ruled).
- FABRIC-3.10.8: Captain Bob, 2026-09-25: "until we close everything, do
not do ant DoE eecution." MC-4's campaigns wait; stage evidence is the
standard sequence plus probes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Captain Bob, 2026-09-25: "pin observr and dynmically utilize worker VM's";
MC-R1..R5 as proposed; MC-R6 changed to start now; O-D relaunches with
row path A (serial-only DoE rows). The amd64 O-D launch was stopped after
81 min / 131,537 rows for this.
- FABRIC-3.10.8: the traced inventory of shared state (kernel subsystems
and the shared VM code -- notably one global dictionary lookup index for
all VMs), and stages MC-0a..MC-4 with evidence named in advance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Design only, for Captain Bob's review; nothing authorizes an edit.
- Observers (the Tripod) get system-core affinity where cores allow;
workers are placed nowhere -- every worker core asks the existing
compudynamic decision at clean boundaries and claims the winner with one
atomic compare-and-swap. Static worker placement ruled out (Captain Bob:
"anti starshipos given there isn't a fixed number of vm's").
- Traces what exists (no SMP bring-up; MADT/PSCI groundwork; no-op kernel
mutexes; single-writer globals), the per-ISA bring-up, the locking
inventory, idle/wake by IPI, per-core physics, staging MC-0..MC-4, and
rulings MC-R1..MC-R6.
- Part II: the open gaps. Filesystem ruled out ("NEVER! anti POSIX!"),
networking deferred; O-D/merge, multi-core, interrupt-driven devices,
the fixed-size audit remainder, security policy, hardware, M8/M9,
FORTH-readable counters, and ten known defects (one new: xHCI input
context leak per configure).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- DOE sets up, JOINs EV-DONE and starts trial 1, then returns. Each trial
births and sends WL-RUN only; completion arrives as the OS's EV-DONE.
- ON-DONE counts this campaign's own workers while active; the last one
ends the trial (KILL all, marker, persist) and starts the next, or
finishes (HB-OFF, UNJOIN, MU-CAMPAIGN: complete). MU-WAIT-ALL is gone;
the start/end cycle goes through MU-END-XT.
- WL-DONE-ARGS removed from all ten workload capsules; lint passes.
Smoke campaign (amd64, 1 rep, rows off per Captain Bob): DOE returns, the
prompt answers mid-trial, all 6 trials chain on events, fail=0 each,
28/28 born/killed/departed, MU-CAMPAIGN: complete in ~61 min. Standard
sequence passes on all three ISAs (with the xHCI fix it exposed, df150e86).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
xhci_cmd_address_device() sized the Output Device Context for Slot + EP0
(64 bytes); xhci_cmd_configure_endpoint() then allocated a bigger copy and
repointed DCBAA[slot_id]. The controller keeps the address it was given at
Address Device (QEMU caches it), so every bulk Endpoint Context was written
past the 64-byte block into the next kmalloc header -- on every USB
mass-storage attach, every ISA. After O-C's smoke campaign reshuffled the
heap, a kfree at thumbdrive unplug walked into it: #GP in coalesce_neighbors().
Found with temporary instrumentation (load base, code bytes at RIP, event
crumbs, a heap-list checker): the corrupt word was an EP Context (Bulk IN,
Running, MPS 1024) exactly DCI 3 x 32 bytes into the 64-byte context.
Now: XHCI_DEVICE_CTX_BYTES (32 contexts x 32 bytes) allocated once per
slot at Address Device, zeroed whole, never reallocated; Configure
Endpoint just fills it. Also ends the per-attach leak of the old context.
Verified: heap checker clean from boot; campaign -> O-A -> MINT -> unplug
-> re-attach passes with 0 exceptions; 3-ISA acceptance passes, parity and
dict_hash set unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- sk_vm_idle_point(may_drain_self): with no frame live and mail pending,
Hera drains one message and returns to the poll loop. Readline passes
n == 0 (the mid-edit gate), headless 1, the key read 0.
- Removed sk_repl_idle()'s timer-paced drain of Hera's queue and the two
flags that guarded it (g_mama_interpreting, g_idle_pump_active) -- the
idle point's frames check already covers that re-entry hazard.
Probe (reverted), all three ISAs: admission -> drain for mail to an idle
Hera 0-1 beats (was up to 100); a message arriving mid-edit is held until
the line is entered (2080-3587 beats), never printed into typed text; an
event queued during a busy line drains when it ends; BLK-ATTACH at boot 1
beat. Acceptance passes; parity, dict_hash set and relay ledger unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Permanent event channels 1/2/3 created at boot beside channel 0, exposed
as FORTH constants EV-DONE / EV-BORN / EV-DIED; CHANNEL-CLOSE refuses 0..3.
- sk_hermes_event(): `S" <name>" ON-...` as type 32 to every member except
the VM it's about; never blocks -- undeliverable copies are dropped,
counted (EV-DROPPED), printed and logged.
- Fired for participants only: EV-DONE from the switch trampoline when real
work runs dry (funded by the VM), EV-BORN at a LIVE VM's first naming
(funded by the newborn), EV-DIED on KILL/reap after departure (funded by
the root). VM.ev_real_work keeps event handling from firing EV-DONE, so
mutual subscribers can't storm.
Console evidence identical on amd64/aarch64/riscv64: one BORN per birth,
one DONE per job run dry (2 queued -> 1), one DIED per KILL, no SEND in any
job text; storm guard: exactly one reaction; channels 1-3 unclosable;
non-joined VMs see nothing; EV-DROPPED 0. Flood probe (reverted): 32
delivered, 7 dropped and counted, never blocked, all heat returned.
dict_hash set changes by the 4 words (546), identical across ISAs;
acceptance passes with held=2048.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
capsule_vm_kill() and capsule_vm_force_reap() never touched kernel-Hermes:
a dead VM stayed a member of channel 0 and every channel it joined, its
undelivered mail was never released, and anything published to it
afterwards held heat forever. M-B's autonomous events would leak on every
dead subscriber. Captain Bob: "Fix first, own commit."
- sk_hermes_vm_departed(): unsubscribe from every channel, release + pop
the pending queue (heat back to each funder, ledger exact), free the
queue entry. Called from both kill and reap, with a HERMES departure line.
Console evidence identical on amd64/aarch64/riscv64: wa with 2 pending ->
"left 2 channel(s), released 2 pending", gone from .CHANNELS, PUBLISH to
its channel delivers 0; acceptance ledger held=2048 (6144 without the
release). Boot parity/dict_hash unchanged; acceptance passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Eleven words in repl.c over the existing kernel-Hermes C API, registered
for every VM; `from` always derived in C. Names as ruled (M-1) except
UNJOIN for LEAVE, which is FORTH-79's DO-loop exit (Captain Bob).
- INJECT (M-2) sends as the root, funded from Hera's reservoir, as new
type 31; pinned STRICT in kernel_main.c beside BIRTH.
- Read-only accessors sk_hermes_pending_at() / sk_hermes_channel_member_at()
for .MAIL / .CHANNELS.
Console evidence identical on amd64/aarch64/riscv64: PUBLISH delivers 2
(both print), LEAVE-VM then PUBLISH delivers 1, INJECT shows in MAIL?/.MAIL
as type 31 from Hera before the recipient runs, CHANNEL-CLOSE succeeds and
refuses channel 0, JOIN/UNJOIN, refusals 0/-1. dict_hash set changes by the
11 words (542 words) and is identical across ISAs; acceptance passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The static sk_hermes_msgs[32] is replaced by a directory of fixed-size
segments (SK_HERMES_ARENA_SEGMENT, growth granularity only). Messages
never move -- callers, including the drain across vm_interpret_kind(),
hold message pointers across later allocations, so copy-on-grow (as
§7.0 first proposed) would dangle them. Each message carries its index;
numbering unchanged.
- Pending-queue rings and channel member lists heap-grown (doubling);
member lists freed on destroy; reassembly scratch sized per call.
SK_HERMES_MSG_MAX / _PENDING_MAX / _MEMBER_MAX are gone.
- Q_SLOT = Q48_ONE / 32 as its own constant (M-0b), value unchanged.
- Every grow failure is printed, logged at LOG_ERROR, counted
(sk_hermes_grow_failures()) and refused whole; a send whose queue can't
grow releases its message.
Probe (reverted), identical on amd64/aarch64/riscv64: 70 messages in one
queue (arena 32 -> 96), sender refused by its reservoir, 72 popped intact
in order, 100 members; a live Hera drain on a full arena grew it mid-drain
and released/popped the right message; each grow path's injected failure
refused whole and counted. 3-ISA acceptance passes; boot parity, dict_hash
set and the live ledger line byte-identical to before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The static word_scratch_addr shared by every VM is gone: WORD builds in
vm->word_buf (257 bytes: count + 255 + NUL). A token over 255 chars is
refused whole (delimiter restored, >IN unconsumed, vm->error), logged at
LOG_ERROR and counted -- no more silent cut to 62 chars.
- Pre-existing corruption found by the probe: TIB/>IN/SPAN (and WORD's
scratch) were allotted lazily at HERE; POST's restore_dict_state(),
FORGET and COLD rolled HERE below them and compiled the dictionary over
them (Hera's TIB lay under ACL-RECHECK's body; QUERY/WORD wrote into
code). Per Captain Bob, all four are now allotted first in both
vm_init()s (offsets 0-1305); COLD's clamp never goes below them.
Probe (reverted), identical on amd64/aarch64/riscv64: 0 colon bodies inside
the buffers; per-VM isolation; 62/100/255-char tokens full length; 300
refused and counted; interpreter sane after full-width TIB writes.
3-ISA acceptance passes; M7.1a hash and dict_hash set unchanged (here
0xd8 -> 0x5f8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- vm_create_word()'s S-1b refusal path now calls transition_metrics_cleanup()
before freeing the metrics struct (its context_window leaked).
- FABRIC-3.9.8: the transition_heat grow is untested -- ENABLE_PIPELINING=0
in the kernel AND the default hosted build; the doc wrongly said hosted.
- Recorded the review: moving id tables have no unlocked or interrupt
readers (heartbeat lookups hold dict_lock, as does the grow); sweep for
other word-id-indexed arrays found none unbounded. Status line updated.
3-ISA acceptance passes; boot parity and dict_hash set unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- word_id_map / recycled_word_ids are heap tables (one shared capacity,
64 doubling), grown all-or-nothing by allocate-copy-free. Numbering is
unchanged, so boot parity is unchanged.
- vm_create_word() takes the id before linking; on failure it frees the
entry, sets vm->error, logs LOG_ERROR and counts
(vm_dictionary_word_id_refusals()). No word runs with WORD_ID_INVALID.
- Readers of the id space bound by vm->next_word_id, not DICTIONARY_SIZE.
- transition_heat gets its own capacity and grows on record (was sized
once -- a heap overflow under a dynamic bound); every reader bounds by it.
- Stadium id->cell rows start empty and grow on dispatch; a failed grow is
logged, counted and not admitted. S-2 (the cell pool) stays open.
Probe (reverted), identical on amd64/aarch64/riscv64: +2000 words get real
ids (capacity 4096); a word at id 2738 lands 200x in the rolling window and
is Stadium-admitted; FORGET recycles across the old boundary; an injected
grow failure refuses the word whole (NULL, error, counter, dictionary
unchanged) and the retry succeeds. 3-ISA acceptance passes; boot parity
(531, 0x572daac83b797332) and the dict_hash set unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- MAX_DICT_ENTRIES (2048) removed: the entry list is allocated to the
dictionary's size; a cyclic chain is detected (Floyd) instead of capped.
- Colon-body walk bounded by vm->here, not a 1024-cell cap.
- Every failure prints PARITY:INCOMPLETE #n with its full reason, logs at
LOG_ERROR, bumps sk_parity_incomplete_count(), and the record reads
hash=INCOMPLETE / PARITY:FAIL code=4 rather than a plausible hash.
- Pre-existing #GP fixed: the walker dereferenced inline body data (S"
bytes, offsets); it now follows only known dictionary entries.
- LIT operands that are entry pointers are hashed by word_id, not address,
so a mature dictionary hashes identically across ISAs.
Probe (reverted): 563 -> 2563 words hashed, identical on amd64/aarch64/
riscv64; cyclic chain -> INCOMPLETE, counter 2. Boot parity unchanged
(531, 0x572daac83b797332). 3-ISA acceptance passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
physics_freeze_words.c, dictionary_heat_diagnostic_words.c,
physics_diagnostic_words.c, physics_benchmark_words.c treated dsp as a
count (data_stack[dsp++] push, data_stack[--dsp] pop) while the VM's dsp
is the top index. On an empty stack a push wrote data_stack[-1], 8 bytes
before the VM struct -- POST's DECAY-RATE@ test did it every boot (on
aarch64 it hit parity.c's counter, elsewhere the POST block RAM tail).
All 28 sites now use vm_pop/vm_push; four one-item guards fixed (dsp<0).
Evidence on amd64/aarch64/riscv64: bytes before the VM struct unchanged
across POST; DECAY-RATE@ leaves 1 item; LOOKUP-STRATEGY! now works;
G interleave, precedence, console under load, relay 3, dict_hash unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Captain Bob: close the messaging gap first; the arena must be dynamic;
audit every fixed-size structure ("no arrays ... it should all be fairly
dynamic"); failures must be handled properly (never silent, never
partial, logged + counted + refused, parity marked INCOMPLETE). All
rulings as proposed; S-2 (Stadium cell pool) left open -- it was a
question, not a proposal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Observer flag on switch-signal slots, set by the kernel for the Tripod
(is_fleet_foundation, hoisted; Hera in kernel_main). The decision: an
observer in progress or holding an event keeps the CPU; observers with
events (mail, Hera's console input) go ahead of participants, oldest wait
stamp first; participants compared by work heat as before. Observers are
never charged (O-2) nor floored. Uncommitted heat-share charge discarded.
Evidence, byte-identical amd64/aarch64/riscv64: G interleave unchanged;
Hestia's mail runs before any worker output (HHHABB...); a line typed
while two workers run is read and executed mid-stream (...ABAB1 2 + . /
3 ok / ABAB...) -- console responsive under load. Relay 3, dict_hash
unchanged. Busy-wait DOE deliberately not run (deadlocks by design until
O-C). FABRIC-3.9.0 records rulings and Captain Bob's messaging note (§7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Captain Bob: Hera, Artemis and Hestia no longer participate in the
calculations; they observe them and act on events. Design (3.9.0):
observer role from is_fleet_foundation, excluded from decision/work
heat/floor, event-driven (mail, console input, device events), Hera's
own mail at the idle point, event-driven DoE driver replacing the busy
wait. Rulings O-1..O-4 + discard of the uncommitted heat-share charge
open. Plan (3.9.8): stages O-A..O-D then merge, evidence named in advance.
FABRIC-3.8 gets a continuation pointer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
logs/20260924-062703 and logs/20260924-081340 plus their run CSVs; they
remain in history at 00058e53. FABRIC-3.8 entry updated to match.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fix belongs in the kernel's compudynamics, not in the DoE capsule
steering its own scheduling. Logs from the stopped relaunch and the check
run stay committed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first amd64 relaunch (stopped by Captain Bob after 1h41m) showed Hera's
busy wait taking an equal CPU share and writing 43% of all heartbeat rows.
100 WAIT charges her 100 beats of work heat per iteration while executing
almost nothing, so workers get the CPU. cfg-0 check: 115 s vs ~210 s,
fail=0, 2/2 dead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each shown dead three ways (FABRIC-3.5 §XXII): the pump loop by its
condition, by construction (birth cap == switch table capacity, every birth
registers) and at run time (probe silent through boot, G interleave, an
8-worker DoE trial and the relay); SWITCH-MARK-WORK by callers, capsules
(source and baked), experiments/tools/logs; the depth counter and back_to
field by having no readers.
Kept: Hera's own pump drain (last timer-paced delivery, Hera only) and the
always-0 readiness CSV columns (a DoE schema call). Acceptance on amd64/
aarch64/riscv64: G interleave unchanged, cfg-0 trial fail=0 2/2 dead,
relay 3, idle growth 0; dict_hash set changed by the removed word
(word_count 531) and identical across ISAs. FABRIC-3.8 stages A-I done.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Workload capsules the DoE births are definitions only; each defines WL-RUN
(its old birth-time work plus its entry word twice) and WL-DONE-ARGS. The
driver births N workers, SENDs each 'WL-RUN WL-DONE-ARGS SEND', waits in an
ordinary loop for N MU-DONE messages, KILLs them (parked idle), and emits
the marker. The serialized VM-EXEC round-robin is gone.
Kernel: no switch checkpoint after a payload's last word; work heat is the
assessed per-beat charge, and a VM gaining work is raised to the busy set's
floor (Captain Bob's ruling) -- fixes a live stall where a busy-waiting Hera
never yielded, without serializing batch starts.
Evidence: cfg-4 rows spread evenly over Hera and all 8 worker slots (44k
switches); full cfg-4 trials on amd64/aarch64/riscv64 end fail=0 with 8/8
KILL dead; Stage G interleave byte-identical on all three. Acceptance:
dict_hash unchanged, relay 3, idle growth 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SEND ( paddr plen name-c name-u -- ok? ): general kernel-Hermes send to a
named VM, from = calling VM, one block max (refused, not truncated), every
VM. New SK_HERMES_MSG_TYPE_SEND (30).
Ruling C-4 (beta), Captain Bob: the switch decision compares work heat --
the heat each VM pulled in by its own per-beat charges -- not absolute
heat, so Hera's structural endowment no longer makes her hottest. Before
it, the two-worker test ran serial (AAAA...BBBB); after it, byte-identical
on amd64/aarch64/riscv64:
AAABBBBBBBBAAAAAAAABBBBBAAAAABBBBBAAAAABBBBBAAAAABBBBBAAAAABBBBBAAAAABBBBBAAAABB
(40 A, 40 B, 16 runs; ordinary words only, nothing yields). Acceptance:
relay 3, idle growth 0; dict_hash set changed by the new word
(word_count 533) and identical across ISAs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
capsule_birth_baby() registers each VM at its LIVE completion point, beside
the common-channel subscription. register() is idempotent. The Artemis/
Hestia registrations in kernel_main.c and WIREBIND's own call are removed
(Hera, never born through capsule_birth_baby, stays with set_current).
Probe (restored away): Tripod registered once each (3 slots), a
WORKER-BIRTH VM becomes a participant with no hand registration (4) and
KILL frees it (3), WIREBIND registers exactly console+user, zero
duplicates, on amd64/aarch64/riscv64. Acceptance: dict_hash unchanged,
relay 3, idle growth 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sk_vm_parked_idle(): saved context + exactly its own IDLE frame.
capsule_vm_kill_refusal() is the single kill rule, shared by
capsule_vm_kill() and the KILL word: refuse SWITCHED_OUT mid-task, parked
with mail pending, or the VM running; allow parked-idle. capsule_vm_kill()
now unregisters the switch slot (G6).
Finding fixed: vm_registry_alloc() never initialized pending_reap (kmalloc
does not zero); C-b's reap_pending walks every participant, so garbage
reaped a healthy VM (seen live before any KILL). Now zeroed at alloc.
Probe (reverted): baseline KILL refused, slots 3->4->4 (FAIL, proves G6);
with Stage D dprobe parked idle, KILL dead, slots 3->4->3, PASS on amd64/
aarch64/riscv64. Acceptance: dict_hash unchanged, relay 3, idle growth 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Who runs next is now sk_vm_switch_decide(): among participants with work
(pending kernel-Hermes mail, parked mid-task, or Hera with console input),
the coolest by VM-physics heat; ties with the incumbent keep the incumbent,
ties among candidates go to the longest waiting (unique wait stamps). No
scheduler state, no timer: heartbeat_tick() no longer touches switching,
readiness/threshold/has_work/take_pending removed. Decided at owned word
checkpoints (colon words and, new, between top-level words) after a beat
or new mail, at the trampoline between messages, and at Hera's idle point
in the console poll loops. The idle pump skips participants (G3 closed for
them): the WIREBIND relay is delivered by a real switch to rajames~user.
Probe (restored away): relay via switch Hera->rajames~user->Hera on all
three ISAs, parked saved_sp inside own stack, walk-order disagreements 0
over tens of millions of decisions, 0 bytes idle log growth. Acceptance:
dict_hash unchanged, relay 3, no faults. Rulings: C-3, C-5 as proposed;
C-4 taken as (alpha), consequence recorded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vm_physics_charge(): per-beat slope into a per-VM carry, whole units pulled
from the rest of the live fleet in proportion to heat (conservative
vm_physics_transfer), rounding shortfall kept in the carry, no fleet_window
sample. Called from vm_tick() only for the switch-current VM. Closes G9:
a VM running its own mail on its own stack now heats up while it works.
Probe (reverted): baseline Hestia heat 0->0->0 over 71 beats (FAIL, proves
G9); with charge 0->11->23 at beats 15->51->86, identical on amd64/aarch64/
riscv64, conserved, PASS. Acceptance: dict_hash unchanged, WIREBIND relay 3.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Captain Bob: "give frames IDs and call by ID" -- frames stack per VM, each
with an ID. New vm_frame.h/.c: VMFrameEntry { VMFrameID id; VMUuid vm_id;
kind (INTERPRET/DRAIN/IDLE) }, a growable VMFrameStack embedded as VM.frames
(reserved at birth, freed in vm_cleanup, grown by allocate-copy-free).
vm_interpret() -> vm_interpret_kind() pushes a frame onto the running stack's
owner, holds only its ID, fails closed if it cannot grow, and pops by ID
(mismatch reported as VM FRAME CORRUPTION).
The Stage A/B loose fields are gone and everything is derived from the stack:
- ownership gate: one frame, and it is vm's own
- drain guard: a DRAIN frame for vm already on the running stack
- parked-idle: an IDLE frame the trampoline pushes around its yield
- log attribution: the top frame's vm_id, resolved by ID through the registry
The single saved-and-restored attribution global is removed.
Evidence: the probe baseline on the pre-frame code read HSSAS (the global
named Hestia while Hera resumed). With frames it reads HSHAS on all three
ISAs, with Hestia's top frame IDLE and Hera's frames back to 0; the probe is
reverted. Formal acceptance on all three ISAs: dict_hash identical and
unchanged, zero UNKNOWN WORD, faults, frame corruption or refused interprets,
and the WIREBIND relay 1 2 + . -> 3 works.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sk_vm_switch_entry() stops being the Stage-2 placeholder bounce and becomes
every switch participant's own idle loop: it drains one of its own
kernel-Hermes messages per pass, on its own native stack (driven by the drain's
return value, so a non-consuming drain cannot spin), and when nothing is
drained it marks the VM parked_idle and yields back. It is entirely OS-driven,
with no VM-side yield word. sk_hermes_drain_checkpoint() gains a per-VM
hermes_draining guard so a payload's own in-word checkpoint cannot drain a
second message inside the first. Still inert in normal operation: nothing
marks work until Stage C.
Finding fixed here: amd64 sk_vm_switch_prime built the first-entry frame at
top-56, entering with rsp 16-aligned instead of the SysV-required 8 mod 16, so
the first aligned SSE spill in a callee took #GP (logs/20260923-142526, kept).
It had been latent since SXXVIII Stage 2 because the placeholder never called
anything. The frame is now at top-64 with a zeroed pad slot. aarch64 and
riscv64 were already correct.
Evidence: a temporary kernel_main probe (reverted, not in this commit) queued
two messages to Hestia, the first calling a colon word, and switched into her.
It passed on all three ISAs: 0 pending, parked_idle, depth 0, context saved on
her own stack, SB-SEQ=12 (drained one at a time, not nested), ledger and
stadium_conserved restored. Formal acceptance without the probe on all three
ISAs: dict_hash identical and unchanged, zero UNKNOWN WORD or faults, and the
live WIREBIND mint -> USE -> 1 2 + . -> 3 relay.
FABRIC-3.8.md also records Captain Bob's rulings that no VM-side word ever gives
up control and that multitasking is a compudynamic hybrid, and opens SS3.12
(VM frames for per-VM execution state).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replaces the global g_vm_interpret_depth test at both execute_colon_word()
checkpoints (kernel-Hermes drain, switch-signal take) with
sk_vm_owns_outermost_frame(vm): the VM physically owns the stack
(sk_vm_switch_current_vm() == vm), is at its own outermost vm_interpret()
frame (new per-VM VM.interpret_depth), and is not compiling. Once VMs
interleave on their own native stacks, a global depth count across all VMs
no longer means "this VM owns this stack"; this is the invariant it was
approximating. The global and the old function stay until Stage I.
Acceptance, all three ISAs, foreground with the framebuffer visible, coupled
disk images reset before each: zero UNKNOWN WORD, zero self-test failures,
dict_hash identical across ISAs and unchanged from the pre-Stage-A boot, and
the live WIREBIND mint -> attach -> USE -> 1 2 + . -> 3 relay with
stadium_conserved(self)=true. Logs: 20260923-140120 (amd64), -140654
(aarch64), -141303 (riscv64).
FABRIC-3.8.md also records Captain Bob's ruling that preemption comes from
the compudynamics, not the timer: principle 5 is corrected and the design
question is opened as SS3.11, to be ruled before Stage C.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The per-ISA amd64 campaign (12345 3 S" amd64" DOE) was stopped ~49 min in,
zero trials complete: its harness drives workers by synchronous VM-EXEC on
Hera's stack, so the concurrency factor counted resident VMs while execution
stayed serialized. Tracing further showed the v2.1.0 reshuffle stops short of
real concurrency: kernel-Hermes never publishes eligibility (no mark_work in
sk_hermes_send_one), the switch trampoline is still the Stage-2 placeholder,
delivery to passive VMs still goes through the idle pump, only Tripod/WIREBIND
VMs are switch participants, and the stack-ownership gate is a single global.
FABRIC-3.8.md records the gap analysis and a staged A-I design, each stage its
own commit and 3-arch acceptance. FABRIC-3.md SXXXV.16 points to it. Logs and
extracted CSVs from today's boots and the stopped run are kept as audit
artifacts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PER-ISA-NREPS was 18, inherited unchanged from local master, but every doc
comment (this file's own header, doe_region.h's own design comment) assumes
N=3 reps/cell -- 18 trials/ISA, 54 total across 3 ISAs. With the actual
constant at 18, MU-EXEC-CAMPAIGN's own math (n-reps * MU-N-CFG, MU-N-CFG=6)
produced 108 trials/ISA, 324 total -- well past the DoE region ring's
256-slot ceiling, silently dropping the oldest ~68 records to FIFO eviction
mid-campaign. Confirmed with Captain Bob: 3 is correct. Fixed.
AMD64-SEED/AARCH64-SEED/RISCV64-SEED/PER-ISA-NREPS are now VARIABLEs, not
CONSTANTs, so they're settable before a run instead of fixed at compile time.
New DOE ( seed reps isa-addr isa-u -- ) word: the common interface for
running any named experiment (not just the three fixed per-ISA ones) through
one entry point -- tags, checks resume count, runs the campaign.
RUN-AMD64-CAMPAIGN/RUN-AARCH64-CAMPAIGN/RUN-RISCV64-CAMPAIGN are now thin
wrappers over it, using the stored per-ISA seed/reps variables.
Live-verified on amd64: both capsules load via EXEC with zero errors, and
DOE itself was exercised directly (4567 3 S" doe-xyz" DOE) -- correctly
tagged, correctly checked DOE-RESUME-COUNT, and launched a real campaign
(confirmed by live per-tick physics output and growing trial state; a full
18-trial run takes the ~25-30 min/ISA CLAUDE.md itself documents, so the run
was stopped once the wiring was confirmed working rather than run to
completion for a syntax check).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Clean boot, all three architectures: word_count=532, zero UNKNOWN WORD,
matching the DoE port's own commit (2d773ca). No source changes -- audit
logs and BLOCK_MAP.md regeneration only.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Local master had 9 unpushed commits building block-backed DoE persistence on
Artemis's own disk (FABRIC-3.md SXXXV.3/.4/.12, designed and built 2026-09-17,
never pushed) that this branch never inherited after the Tripod/kernel
reshuffle diverged history. Ports the mechanism cleanly onto the current
Hera/Artemis/Hestia shape:
- doe_region.h/.c: growable, CRC-protected trial-summary ring in Artemis's
top-of-device metadata fence, immediately past log_region.c's own ceiling
(devblock_from_top 98-102). Same accessor/CRC discipline as log_region.c.
- log_words.c: DOE-PERSIST-TRIAL and DOE-RESUME-COUNT kernel-only FORTH words.
- multiuser-doe.4th: per-ISA tagging (MU-ISA-TAG!) and campaign resume support
(MU-EXEC-CAMPAIGN now takes a start-idx).
- per-isa-doe.4th (new): RUN-AMD64-CAMPAIGN/RUN-AARCH64-CAMPAIGN/
RUN-RISCV64-CAMPAIGN entry points, resuming from DOE-RESUME-COUNT.
- extract_doe_region.py (new): host-side reader, CRC-verified, no filesystem
assumptions -- reads the ring directly off a raw disk image.
Deliberately not ported: the R/LaTeX report-generation pipeline
(analyse_multiuser_doe.R, multiuser_doe_report.tex, build_multiuser_doe_report.sh)
-- not needed to prove persistence works, and unscoped for this pass.
Verified live on all three architectures: clean boot, word_count 530->532,
zero UNKNOWN WORD. amd64 and riscv64 both live-exercised end-to-end (write via
DOE-PERSIST-TRIAL, count via DOE-RESUME-COUNT, clean QEMU exit, then read back
correctly with extract_doe_region.py against the raw disk image -- no host
serial dependency, matching the real bare-metal requirement this campaign is
rehearsing for).
riscv64 finding (not a defect in this port, no fix applied): DOE-PERSIST-TRIAL
failed 3/3 on a first attempt against a fresh disk image with
"virtio_blk: vblk_io timed out". Root cause: disk/artemis.img is a sparse file
(180KB allocated of a 1GB apparent size); the new DOE region (devblock_from_top
98-102) is genuinely virgin territory on a fresh image, and riscv64's slower
TCG emulation missed the existing virtio_blk spin-bound budget (already
documented as fragile, FABRIC-3.md 2026-09-09) on that first cold touch.
Confirmed by isolation: a plain BLOCK/UPDATE at an already-warm LBN succeeded
instantly, the pre-existing LOG-ERROR" (same meta-zone mechanism, an
already-warmed offset range) also succeeded, and a reboot reusing the
now-partially-warmed disk let DOE-PERSIST-TRIAL succeed immediately and
verify correctly on read-back. Pre-existing environmental fragility, not a
logic bug in the ported address computation.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Distinguishes it clearly from the still-accepted "cloned outside
StarshipOS entirely" limitation this document already settled -- Phase
8 v3 closes a narrower, different threat: cloning block content using
StarshipOS's own console primitives, now refused by MOVE/CMOVE/CMOVE>/
RELOCATE-BLOCK for cross-device copies.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two research passes confirmed the identity record itself (seed/pubkey/
cert) is already unreachable from any FORTH primitive -- only C-level
read_devblock/write_devblock touch it. But a device's ordinary user block
content CAN be copied between two attached devices today, using only
stock, unpinned words: <src> BLOCK <dst> BUFFER 1024 MOVE UPDATE
SAVE-BUFFERS, or the dedicated RELOCATE-BLOCK word (whose own doc comment
already admits "performs no policy validation of its own"). Checked
whether the existing per-block owner_fp/BLK-ACL-ALLOW@ metadata already
solves this -- it doesn't: owner_fp encodes who (a VM identity pubkey),
never where (physical device), and blk_get_buffer()/blk_update() never
consult acl_allow/acl_ttl at all -- those fields are completely inert.
Small, targeted fix, no rearchitecture:
- blk_subsys_relocate_block() (block_subsystem.c): same-device check.
Its own documented purpose is wear-leveling (relocate on the SAME
device) -- never stated as cross-device, and nothing enforced that
until now.
- New public blk_lbn_device_handle() (block_subsystem.c/.h): the missing
LBN-to-device direction (blk_get_device_range() already goes the other
way). Opaque, stable, == comparable.
- MOVE (memory_words.c) and CMOVE/CMOVE> (string_words.c): refuse when
both addresses are block-window addresses backed by two different
devices -- the exact shape of the composed attack. A copy where either
end is ordinary VM memory (the overwhelming common case: staging text
from PAD, editing a block in place) is untouched.
- blk_vm_check_epoch()/blk_vm_slot_for_addr() exposed (block_words.h) so
the two new call sites share the same window-slot invalidation contract
rather than a second, divergent copy of it.
Verified live on all three architectures, not just boot-clean: same-
device MOVE/RELOCATE-BLOCK still succeed exactly as before; cross-device
MOVE/CMOVE/RELOCATE-BLOCK all refused. Zero UNKNOWN WORD, identical
dict_hash across all three (this change adds no FORTH-visible word, only
internal refusal conditions, as predicted).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Captain Bob rejected a PIN/passphrase second factor firmly and directly
after it was built and live-tested on all three architectures: "nothing
like a pin or a password or secret code or any bullshit... Everybody has
secrets. There's only the drive." All PIN-related code (KDF, XOR
keystream seed encryption, no-echo input, MINT/WIREBIND prompts,
user_identity_seed_t v3 format) was reverted before commit -- none of it
ever landed in git history.
This project's identity model has no knowledge factor, by design:
physical possession of the thumbdrive is the entire credential. A
byte-for-byte clone being equivalent to the real drive is the accepted
model, not a gap needing a fix. Recorded here so future work doesn't
default back to a PIN/password approach.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Found while starting Part A's implementation, before any code was written
against the original design: reading the actual deleted SEND-ELEVATE-REQUEST
source (git show 3e201c8^:capsules/common/messaging.4th, block 5040) shows
it copied the target word's name as literal character bytes into a scratch
buffer, building "S" <name-text>" <pk0> <pk1> <pk2> <pk3> ELEVATE-GRANT"
entirely in the sending VM's own memory, then sent that finished string --
never a raw address -- to Hera. waddr/wu never crossed the VM boundary as
numbers anywhere in this flow. The original write-up reasoned from
ELEVATE-GRANT's own signature alone, without first reading how the caller
actually built its message.
Corrected in place, wrong original text kept struck-through for
traceability rather than deleted, per this series' own convention.
Net effect: Part A (the buffer/message redesign) is not needed -- the
original mechanism was already safe. Part B (capsules/zuse.4th, already
committed and three-arch verified this session, 089ab21) stands on its
own, unaffected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
ZUSE-ELIGIBILITY-ADD's own doc comment admitted "no authorization check
here or anywhere else... applied later if and when actually needed --
not invented here." That's now: anyone reaching a Hera FORTH prompt
could add their own pubkey to the eligibility list with zero legitimate
identity material -- no minted drive, no WIREBIND, no cert-signature
check involved at all. Once a future caller reaches ELEVATE-GRANT again,
a self-added pubkey would pass zuse_eligibility_is_member() and grant
ACL-ALLOW!/ACL-TTL! on any named word.
Fixed the FORTH-only way, matching this project's own convention (ACL
policy belongs in ACL.4th, never in C; never gate on zuse_session in C --
her power is the absence of ACLs, not a hardcoded session check):
ZUSE-ELIGIBILITY-ADD is now denied by default (capsules/zuse.4th block
4016), granted and pinned only inside ACL-ZUSE-BOOT's already-existing
authenticated branch (block 4017) -- the same gate her own god-mode
already goes through, requiring a real cert-verified Zuse before it opens.
Live-verified on all three architectures, not just boot-clean: after
genesis authentication, ACL-ALLOW@ and ACL-PINNED? both read -1, and
HERE ZUSE-ELIGIBILITY-ADD executes successfully past the ACL gate.
Phase 8 v1 plan: /home/rajames/.claude/plans/jiggly-cuddling-stallman.md
Part A (the ELEVATE-GRANT pointer-confusion fix, FABRIC-3.7.md) is
separate, not yet built.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The hosted Makefile writes include/version.h with FORCE as a prerequisite
(always regenerates), but Makefile.starkernel's own rule had no
prerequisite at all -- Make only rebuilds a target with no prerequisites
when the file is missing. Since both Makefiles write the same path with
incompatible content (the hosted version has no LITHOS_VERSION/
LITHOS_VERSION_STR at all), running a bare `make -f Makefile.starkernel`
after a hosted `make` build silently reused the wrong file and failed
deep in kernel_main.c with "LITHOS_VERSION_STR undeclared".
Added FORCE (declared .PHONY, matching the hosted Makefile's own existing
pattern) as include/version.h's prerequisite in Makefile.starkernel.
Verified the fix directly: poisoned version.h with a hosted build, then
ran a bare (non-clean) kernel build and confirmed it self-heals.
Full three-architecture acceptance: amd64 (logs/20260922-215333/),
aarch64 (logs/20260922-215803/), riscv64 (logs/20260922-220217/) -- all
three reach [zuse@Hera] ok>, zero UNKNOWN WORD.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
New document, not a reopening of the closed FABRIC-3.5.md/FABRIC-3.6.md --
successor for exactly one topic, per those documents' own close discipline.
Records a security defect found by inspection while auditing Phase 4's
collateral damage to ELEVATE-GRANT: the old SEND-ELEVATE-REQUEST mechanism
passed a raw address (waddr/wu) computed in the sending VM's own memory
space across to Hera, which dereferences it in Hera's own space --
per-VM vaddr_t means those are never the same address space. Whoever
controls waddr controls what dictionary entry NAME>XT resolves to on
Hera, independent of the caller's actual pubkey/eligibility.
Design fix: never cross an address, only ever cross bytes -- generalizes
this session's own payload-aliasing fix (SkHermesMessage.payload_buf) one
level up. Send the target word's name as inline payload bytes, copy them
into a fixed kernel-owned buffer already in Hera's own memory on receipt,
and hand vm_interpret() only kernel-controlled integer literals referencing
that buffer. ELEVATE-GRANT itself is unchanged -- policy logic stays in
FORTH, per ACL.4th's own rule.
No code written or authorized by this document.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
