rajames/LithosAnanke
1
1
Fork 0
Code Issues 61 Pull Requests Actions Packages Projects Releases Activity

fix(hermes): one delivery path -- a VM takes its own messages on its own turn, never on another's stack

Browse Source 
FABRIC-3.5 SXXXII, closing FABRIC-3 SXXIX bug 2 (SXXXVI).  A send marks the
receiver for the switch signal; the VM's own loop on its native stack takes
its messages and yields; Hera's idle beat makes the pending switch instead
of running other VMs' messages on her stack; every VM gets switcher turns
from birth; the drain refuses any call outside the VM's own context (the
test: it held every relayed line before, none now).

Found on the way: amd64's first-entry frame left rsp 8 bytes off the SysV
entry alignment (#GP on the first interpreter run on a VM's own stack), and
a VM yielding from its loop was marked SWITCHED_OUT, which stopped console
relay after one message and would have made KILL refuse it.

Relay of four lines to a WIREBIND identity on all three ISAs, five
hot-plugged identities, and standard acceptance (POST 1052, hash
0x08873e0f44b7cb2a) all pass.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
rajames
2026-10-14 10:37:57 -04:00
co-authored by Claude Opus 5.5
parent b7e8e4b058
commit 029a8ed9cf
34 changed files with 110307 additions and 119 deletions
Download Patch File Download Diff File Expand all files Collapse all files
kernel/src/vm/switch.c
+74 -24
View File
@@ -30,6 +30,8 @@
#include "starkernel/vm/switch.h"
#include "vm.h"
#include "starkernel/capsule_birth.h" /* capsule_vm_set_state */
#include "starkernel/capsule_vm_switch_signal.h" /* one path: take_pending(), mark_work() */
#include "starkernel/vm/kernel_hermes.h" /* one path: sk_hermes_drain_checkpoint() */
#include <stdint.h>
/* Arch-specific (switch.S in each arch dir). sk_vm_switch_prime() builds
@@ -51,6 +53,8 @@ extern uint64_t sk_vm_switch_prime(uint64_t stack_top, void *entry);
* who most recently switched into it does. */
static VM *g_switch_entry_vm;
static int switch_core(struct VM *from, struct VM *to, VMState from_state);
/* FABRIC-3.md §XXVIII Stage 3 follow-on (2026-09-14): which VM is
* currently running, from the switch mechanism's own point of view --
* independent of vm_log_attributed_vm() (vm_core.c), which is scoped to
@@ -72,39 +76,81 @@ VM *sk_vm_switch_current_vm(void) { return g_switch_current_vm; }
void sk_vm_switch_set_current(VM *vm) { g_switch_current_vm = vm; }
/**
* sk_vm_switch_entry - trampoline for a VM's first-ever entry.
* sk_vm_switch_entry - what a VM runs on its own native stack.
*
* A freshly-entered VM has no production behavior defined yet -- what a
* switched-to VM should actually DO once running is later-stage work
* (message dispatch, an idle loop, whatever Stage 3+ needs). For now this
* is a minimal, permanent placeholder: immediately yield back to whoever
* most recently switched to it, forever. Reads self->native_switch_back_to
* Entered once, at a VM's first-ever switch-in, and never left: each turn
* it takes its own pending kernel-Hermes messages, then yields back to
* whoever most recently switched to it. Reads self->native_switch_back_to
* fresh every iteration (not a value captured once) so a later switch-in
* from a different VM is honored, not silently bounced to whoever entered
* first. Must never fall off the end -- there is no legitimate return
* address below a synthetic first-entry frame, unlike an ordinary
* function.
* from a different VM is honored. Must never fall off the end -- there is
* no legitimate return address below a synthetic first-entry frame.
*
* FABRIC-3.md §XXVIII Stage 3 follow-on (2026-09-14): an attempt to make
* this attributable by calling vm_interpret(self, "MSG-TICK") here was
* tried and reverted -- it introduced a second, distinct hang (a stall in
* sk_repl_headless_wait()'s WIREBIND-attachment wait, not root-caused) in
* place of the switch-storm it was meant to fix. The switch-storm's real
* cause (sk_vm_switch_signal_tick()'s "who is current" check via
* vm_log_attributed_vm() never seeing a VM that only ever runs this raw
* C trampoline) is instead fixed at the source in switch.c/
* capsule_vm_switch_signal.c: sk_vm_context_switch() now tracks "current"
* itself, independent of vm_interpret()'s own attribution bookkeeping.
* This placeholder's behavior is therefore unchanged from Stage 2.
* History. Until 2026-10-14 this only yielded (a Stage 2 placeholder), and
* a VM's messages were run for it from Hera's idle beat, on her stack --
* the second of the two delivery paths FABRIC-3.5.md §XXXII ruled down to
* one. A first attempt at giving it work (2026-09-14: vm_interpret(self,
* "MSG-TICK") here) was reverted after a stall in
* sk_repl_headless_wait()'s WIREBIND wait, never root-caused; the switch
* storm it was aimed at was fixed at its source instead (switch.c and
* capsule_vm_switch_signal.c track "current" themselves).
*/
void sk_vm_switch_entry(void) {
VM *self = g_switch_entry_vm;
for (;;) {
sk_vm_context_switch(self, self->native_switch_back_to);
/* ONE PATH (FABRIC-3.5.md §XXXII, built 2026-10-14): what a VM does
* on its turn is take its own messages, here, on its own stack --
* the only place a VM that nothing else drives ever takes one
* (Hera's idle beat used to run them for it, on her stack). Not
* while its interpreter is mid-line on some other, parked, stack
* (a console VM waiting in KEY, say): then they wait, and it asks
* for another turn. */
if (self->interpret_depth == 0) {
while (sk_hermes_drain_checkpoint(self) != 0) { }
}
if (sk_hermes_pending_count(self->stadium_vm_id) > 0) {
sk_vm_switch_signal_mark_work(self->stadium_vm_id);
}
switch_core(self, self->native_switch_back_to, VM_STATE_LIVE);
}
}
int sk_vm_context_switch(VM *from, VM *to) {
int sk_vm_switch_service_pending(VM *from) {
VMRegistryEntry target;
VMUuid target_id = sk_vm_switch_signal_take_pending();
if (!from || vm_uuid_equal(target_id, vm_uuid_none())) return 0;
if (capsule_vm_registry_get(target_id, &target) != 0 ||
!target.vm_ptr || target.vm_ptr == (void *)from) {
return 0;
}
/* FABRIC-3.md §XXVIII Stage 4 (2026-09-14): a target marked
* pending_reap is a WIREBIND identity whose owning device has
* physically detached while it sat parked SWITCHED_OUT --
* capsule_vm_kill() already refused to free it when that happened
* (see VMRegistryEntry.pending_reap's own doc comment). This is the
* safe point the mark-and-defer design calls for: reap instead of
* resuming, and skip the switch -- take_pending() already consumed the
* pending flag, so no other caller will try this same dead end. */
if (target.pending_reap) {
capsule_vm_force_reap(target_id);
return 0;
}
sk_vm_switch_signal_note_switch_performed(target_id);
sk_vm_context_switch(from, (VM *)target.vm_ptr);
/* Resumes here once something later switches back. */
return 1;
}
/* `from_state` is what `from` is while it is parked. VM_STATE_SWITCHED_OUT
* for a VM switched out mid-line (the colon-word checkpoint, Hera's idle
* beat): its stack holds live frames, and capsule_vm_kill() must refuse
* it. VM_STATE_LIVE for a VM yielding from the top of its own loop in
* sk_vm_switch_entry() (2026-10-14): its stack holds nothing but that loop,
* so it is as alive, as killable and as reachable by name as a VM that has
* never had a turn -- marking it SWITCHED_OUT made the console stop
* relaying to a WIREBIND identity after its first message, and made KILL
* refuse every VM that had ever taken one. */
static int switch_core(VM *from, VM *to, VMState from_state) {
uint64_t new_sp;
if (!from || !to || to->native_stack_top == 0) {
@@ -126,7 +172,7 @@ int sk_vm_context_switch(VM *from, VM *to) {
}
capsule_vm_set_state(to->stadium_vm_id, VM_STATE_LIVE);
capsule_vm_set_state(from->stadium_vm_id, VM_STATE_SWITCHED_OUT);
capsule_vm_set_state(from->stadium_vm_id, from_state);
g_switch_current_vm = to;
sk_vm_switch_to(&from->native_stack_saved_sp, new_sp);
@@ -138,4 +184,8 @@ int sk_vm_context_switch(VM *from, VM *to) {
return 0;
}
int sk_vm_context_switch(VM *from, VM *to) {
return switch_core(from, to, VM_STATE_SWITCHED_OUT);
}
#endif /* __STARKERNEL__ */