[Audit] --audit-r3-dump-bytes: dump N bytes at r3 when probe fires

AUDIT-059 round 15 — diagnostic. When `--audit-r3-dump-bytes=N` is set,
every `--audit-pc-probe-hex` fire emits a paired `AUDIT-R3-DUMP` line
with N bytes of guest memory from r3 as u32 lanes (4-byte aligned, cap
256B). Sized for the 80-byte stack-local struct at sub_82452DC0's
`r31+96` (probe sub_8245B000 entry where r3 IS the struct ptr).

Settable via `XENIA_AUDIT_R3_DUMP_BYTES` env. Read-only; lockstep digest
unaffected (empty-set fast path in fire_audit_pc_probe_if_match).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

crates/xenia-app/src/main.rs

@@ -242,6 +242,44 @@ enum Commands {
         /// line). Stdout when omitted.
         #[arg(long)]
         lr_trace_out: Option<String>,
+        /// AUDIT-2BF — comma-separated list of guest PCs (hex, no `0x`
+        /// prefix required) to capture as one-line `AUDIT-PC-PROBE`
+        /// records on every fire. Designed for the silph init chain
+        /// virtual-dispatch site at `sub_82172BA0+0x1E8` (PC
+        /// `0x82172D88`, a `bctrl` after a 3-deep vtable-slot-6 load).
+        /// Each record carries (pc, tid, hw, cycle, lr, r3, r11) plus
+        /// four guest-memory dereferences off r3: `[r3+0]` (vtable),
+        /// `[[r3+0]+24]` (slot 6 method = bctrl target), `[r3+0x0C]`
+        /// (auxiliary handle), `[r3+0x30]` (embedded sub-object vtable).
+        /// Compares directly against canary's round-9 capture:
+        /// r3=0xBCCC52C0, [r3+0]=0x820A3644, slot6=sub_821B55D8,
+        /// [r3+0xC]=0xF80000D8, [r3+0x30]=0x820A1870. Read-only;
+        /// lockstep digest unaffected. Settable via
+        /// `XENIA_AUDIT_PC_PROBE`. Example:
+        /// `--audit-pc-probe-hex=82172D88,82172D80`.
+        #[arg(long)]
+        audit_pc_probe_hex: Option<String>,
+        /// AUDIT-2BF round 14 — guest VA (hex, optional `0x` prefix) to
+        /// dereference 3 deep on every `--audit-pc-probe-hex` fire.
+        /// Emits a paired `AUDIT-MEM-READ` line with the singleton value,
+        /// vtable, vtable[0] (= first virtual method, the bctrl target
+        /// at `0x822F1B4C`), and vtable[24] (= slot 6 = canary's silph
+        /// chain target `sub_821B55D8`). Compare ours vs canary to
+        /// determine whether the bctrl dispatches to the same function
+        /// or a different one. Read-only; lockstep digest unaffected.
+        /// Settable via `XENIA_AUDIT_MEM_READ`. Example:
+        /// `--audit-mem-read-hex=828E1F08`.
+        #[arg(long)]
+        audit_mem_read_hex: Option<String>,
+        /// AUDIT-052 — number of bytes (4-byte aligned, max 256) to
+        /// dump from `r3` on every `--audit-pc-probe-hex` fire. Emits a
+        /// paired `AUDIT-R3-DUMP` line with the u32 lanes. Designed for
+        /// the 80-byte stack-local struct at `sub_82452DC0` (`r31+96`)
+        /// when probing `sub_8245B000` entry — where `r3` IS the struct
+        /// pointer. Read-only; lockstep digest unaffected. Settable via
+        /// `XENIA_AUDIT_R3_DUMP_BYTES`. Example: `--audit-r3-dump-bytes=80`.
+        #[arg(long)]
+        audit_r3_dump_bytes: Option<u32>,
     },
     /// Browse XISO disc image contents
     Browse {
@@ -405,6 +443,9 @@ fn main() -> Result<()> {
             probe_db,
             lr_trace,
             lr_trace_out,
+            audit_pc_probe_hex,
+            audit_mem_read_hex,
+            audit_r3_dump_bytes,
         } => cmd_exec(
             &path,
             max_instructions,
@@ -431,6 +472,9 @@ fn main() -> Result<()> {
             probe_db.as_deref(),
             lr_trace.as_deref(),
             lr_trace_out.as_deref(),
+            audit_pc_probe_hex.as_deref(),
+            audit_mem_read_hex.as_deref(),
+            audit_r3_dump_bytes,
         ),
         Commands::Browse { path } => cmd_browse(&path),
         Commands::Info { path } => cmd_info(&path),
@@ -662,6 +706,9 @@ fn cmd_exec(
     probe_db: Option<&str>,
     lr_trace: Option<&str>,
     lr_trace_out: Option<&str>,
+    audit_pc_probe_hex: Option<&str>,
+    audit_mem_read_hex: Option<&str>,
+    audit_r3_dump_bytes: Option<u32>,
 ) -> Result<()> {
     cmd_exec_inner(
         path,
@@ -689,6 +736,9 @@ fn cmd_exec(
         probe_db,
         lr_trace,
         lr_trace_out,
+        audit_pc_probe_hex,
+        audit_mem_read_hex,
+        audit_r3_dump_bytes,
         None,
         None,
         false,
@@ -735,6 +785,9 @@ fn cmd_check(
         None, // probe_db — same
         None, // lr_trace — same
         None, // lr_trace_out — same
+        None, // audit_pc_probe_hex — diagnostic, never wanted on goldens
+        None, // audit_mem_read_hex — same
+        None, // audit_r3_dump_bytes — same
         out,
         expect,
         stable_digest,
@@ -767,6 +820,9 @@ fn cmd_exec_inner(
     probe_db: Option<&str>,
     lr_trace: Option<&str>,
     lr_trace_out: Option<&str>,
+    audit_pc_probe_hex: Option<&str>,
+    audit_mem_read_hex: Option<&str>,
+    audit_r3_dump_bytes: Option<u32>,
     digest_out: Option<&str>,
     digest_expect: Option<&str>,
     stable_digest: bool,
@@ -1167,6 +1223,84 @@ fn cmd_exec_inner(
         }
     }
 
+    // AUDIT-2BF — `--audit-pc-probe-hex=82172D88,...`. Bare-hex tokens
+    // (with or without `0x` prefix). Parses every comma-separated entry
+    // as a u32 PC and inserts into `kernel.audit_pc_probe_pcs`. Empty
+    // set is the hot-path no-op (single is_empty() check).
+    let audit_pc_probe_combined: Option<String> = match (
+        audit_pc_probe_hex, std::env::var("XENIA_AUDIT_PC_PROBE").ok(),
+    ) {
+        (Some(s), _) => Some(s.to_string()),
+        (None, Some(s)) if !s.is_empty() => Some(s),
+        _ => None,
+    };
+    if let Some(list) = audit_pc_probe_combined {
+        for token in list.split(',').map(str::trim).filter(|s| !s.is_empty()) {
+            let hex = token.strip_prefix("0x").or_else(|| token.strip_prefix("0X")).unwrap_or(token);
+            let pc = u32::from_str_radix(hex, 16)
+                .map_err(|e| anyhow::anyhow!("--audit-pc-probe-hex {token:?}: {e}"))?;
+            kernel.audit_pc_probe_pcs.insert(pc);
+        }
+        if !quiet && !kernel.audit_pc_probe_pcs.is_empty() {
+            let mut pcs: Vec<u32> = kernel.audit_pc_probe_pcs.iter().copied().collect();
+            pcs.sort_unstable();
+            let strs: Vec<String> = pcs.iter().map(|p| format!("{p:#010x}")).collect();
+            tracing::info!(
+                "audit-pc-probe armed: {} ({})",
+                kernel.audit_pc_probe_pcs.len(),
+                strs.join(", "),
+            );
+        }
+    }
+
+    // AUDIT-2BF round 14 — `--audit-mem-read-hex=828E1F08`. Single
+    // hex VA (optional `0x` prefix). Stored on `kernel.audit_mem_read_addr`.
+    // Paired with `audit_pc_probe_pcs`: on every probe fire, the kernel
+    // emits a second `AUDIT-MEM-READ` line dereferencing 3 deep so we can
+    // resolve vtable[0] / vtable[24] at the singleton.
+    let audit_mem_read_combined: Option<String> = match (
+        audit_mem_read_hex, std::env::var("XENIA_AUDIT_MEM_READ").ok(),
+    ) {
+        (Some(s), _) => Some(s.to_string()),
+        (None, Some(s)) if !s.is_empty() => Some(s),
+        _ => None,
+    };
+    if let Some(tok) = audit_mem_read_combined {
+        let tok = tok.trim();
+        if !tok.is_empty() {
+            let hex = tok.strip_prefix("0x").or_else(|| tok.strip_prefix("0X")).unwrap_or(tok);
+            let addr = u32::from_str_radix(hex, 16)
+                .map_err(|e| anyhow::anyhow!("--audit-mem-read-hex {tok:?}: {e}"))?;
+            kernel.audit_mem_read_addr = Some(addr);
+            if !quiet {
+                tracing::info!("audit-mem-read armed: {:#010x}", addr);
+            }
+        }
+    }
+
+    // AUDIT-052 — `--audit-r3-dump-bytes=80`. When set, every
+    // `--audit-pc-probe-hex` fire emits a paired `AUDIT-R3-DUMP` line
+    // with N bytes from `r3` (4-byte aligned, capped at 256). Sized for
+    // the 80-byte stack-local struct at `sub_82452DC0`'s `r31+96` —
+    // probe `sub_8245B000` entry where `r3 == parent's r31+96`.
+    let audit_r3_dump_combined: Option<u32> = match (
+        audit_r3_dump_bytes, std::env::var("XENIA_AUDIT_R3_DUMP_BYTES").ok(),
+    ) {
+        (Some(n), _) => Some(n),
+        (None, Some(s)) if !s.is_empty() => Some(
+            s.parse::<u32>().map_err(|e| anyhow::anyhow!("--audit-r3-dump-bytes {s:?}: {e}"))?,
+        ),
+        _ => None,
+    };
+    if let Some(n) = audit_r3_dump_combined {
+        if n > 0 {
+            kernel.audit_r3_dump_bytes = Some(n);
+            if !quiet {
+                tracing::info!("audit-r3-dump armed: {} bytes", n);
+            }
+        }
+    }
+
     // Diagnostic. Parse `--dump-addr=0x828F3D08,...` (or
     // `XENIA_DUMP_ADDR=...`) into `kernel.dump_addrs`. The contents
     // are dumped at end-of-run by `dump_thread_diagnostic`. Pure
@@ -1990,7 +2124,13 @@ fn coord_pre_round(
     }
 
     kernel.fire_due_timers();
-    try_inject_graphics_interrupt(kernel);
+    // Graphics-interrupt delivery is no longer done here — see
+    // `dispatch_graphics_interrupts`, called from the outer loop with
+    // `mem` and `&mut stats` in scope. The audio path still uses the
+    // asynchronous LR-sentinel inject because each XAudio client has a
+    // dedicated worker thread (audit-048 Plan B) that the callback
+    // runs on; we just queue the source and the worker_prologue's
+    // halt-sentinel restore path closes the loop.
     if kernel.xaudio_tick_enabled {
         try_inject_audio_callback(kernel);
     }
@@ -2010,6 +2150,24 @@ fn coord_idle_advance(
     shutdown: &Option<std::sync::Arc<std::sync::atomic::AtomicBool>>,
     stats: &ExecStats,
 ) -> RoundCtl {
+    // Path β (iterate-2.BE follow-up): when the scheduler has no Ready
+    // threads, `coord_pre_round`'s instruction-count vsync ticker stops
+    // advancing (instruction_count is frozen). That starves the
+    // host-driven graphics ISR dispatcher: queue stays empty, no
+    // deliveries occur, and the very stall we're trying to break out of
+    // gets worse. Tick vsync from wallclock here unconditionally — it's
+    // a host-clock read, independent of instruction count, and the
+    // dispatcher in the outer loop will drain whatever we queue on the
+    // next pass. Mirrors the `--parallel` ticker choice in
+    // `coord_pre_round` (`tick_vsync_wallclock` branch).
+    if kernel.interrupts.tick_vsync_wallclock() {
+        use std::sync::atomic::Ordering;
+        let mmio = kernel.gpu.mmio();
+        let prev = mmio.d1mode_vblank_vline_status.load(Ordering::Relaxed);
+        mmio.d1mode_vblank_vline_status
+            .store(prev | 0x1, Ordering::Relaxed);
+    }
+
     let next_timer = kernel.earliest_timer_deadline();
     let next_wait = kernel.scheduler.earliest_wait_deadline();
     let target = match (next_timer, next_wait) {
@@ -2218,6 +2376,7 @@ fn worker_prologue(
     // the helper, no overhead on the hot path.
     kernel.fire_ctor_probe_if_match(hw_id, mem);
     kernel.fire_branch_probe_if_match(hw_id);
+    kernel.fire_audit_pc_probe_if_match(hw_id, mem);
     kernel.fire_lr_trace_if_match(hw_id);
 
     if mem.has_mem_watch() {
@@ -2595,12 +2754,21 @@ fn run_execution(
     let mut workers: [WorkerCtx; xenia_cpu::scheduler::HW_THREAD_COUNT] =
         std::array::from_fn(|i| WorkerCtx::new(i as u8, force_per_instr));
 
+    // Iterate-2.BE — decode cache used by the synchronous ISR
+    // dispatcher. ISRs are short (~40 PPC instructions) but fire
+    // every ~16.7 ms, so persisting the cache across calls avoids
+    // re-decoding the same handful of pages 60×/s.
+    let mut isr_decode_cache = xenia_cpu::decoder::DecodeCache::new();
+
     'outer: loop {
         // Per-round prologue: budget / shutdown / heartbeat / vsync /
-        // timers / graphics-interrupt injection. Carved into
+        // timers / audio-interrupt injection. Carved into
         // `coord_pre_round` so the parallel scheduler (Step 03+) can
         // call the same coordination logic between phaser barriers
-        // without duplicating it from the lockstep path.
+        // without duplicating it from the lockstep path. The
+        // graphics-interrupt dispatch is hoisted out — it runs
+        // *synchronously* (host-driven, iterate-2.BE) and needs `mem`
+        // + `&mut stats` which aren't in `coord_pre_round`'s scope.
         match coord_pre_round(
             kernel,
             &stats,
@@ -2612,6 +2780,13 @@ fn run_execution(
             RoundCtl::BreakOuter => break,
             RoundCtl::Continue => {}
         }
+        dispatch_graphics_interrupts(
+            kernel,
+            mem,
+            &mut stats,
+            &mut isr_decode_cache,
+            thunk_map,
+        );
 
         // Snapshot round schedule. `round_schedule` also advances rng state
         // when seeded; mutation is intentional.
@@ -2789,6 +2964,10 @@ fn run_execution_parallel(
 
     let throttle_start = Instant::now();
 
+    // Iterate-2.BE — decode cache for the synchronous ISR dispatcher.
+    // Lives on the coordinator (this) thread; workers never touch it.
+    let mut isr_decode_cache = xenia_cpu::decoder::DecodeCache::new();
+
     const COORD_ID: u8 = xenia_cpu::scheduler::HW_THREAD_COUNT as u8; // = 6
     const PARTY_COUNT: u32 = xenia_cpu::scheduler::HW_THREAD_COUNT as u32 + 1;
 
@@ -3025,6 +3204,22 @@ fn run_execution_parallel(
             }
             let mut guard = pre_outcome.1;
 
+            // Iterate-2.BE — host-driven synchronous ISR dispatch.
+            // Runs under the kernel lock while workers are still parked
+            // at the phaser B2 barrier (the coordinator hasn't published
+            // the runnable mask or arrived at the phaser yet), so no
+            // contention with worker steps.
+            {
+                let mut s = stats_mtx.lock().expect("stats mutex poisoned");
+                dispatch_graphics_interrupts(
+                    &mut *guard,
+                    mem,
+                    &mut *s,
+                    &mut isr_decode_cache,
+                    thunk_map,
+                );
+            }
+
             guard.scheduler.begin_round();
             let order = guard.scheduler.round_schedule();
 
@@ -3140,146 +3335,275 @@ fn run_execution_parallel(
     stats_mtx.into_inner().expect("stats mutex poisoned")
 }
 
-/// First-Pixels M2 — inject a queued graphics interrupt into HW thread 0
-/// when it's safe to do so (callback registered, no interrupt already
-/// running). Called at the top of each scheduler round.
+/// Iterate-2.BE — host-driven synchronous dispatch of all queued
+/// graphics interrupts. Mirrors canary's
+/// [`EmulateCPInterruptDPC`](../../../../xenia-canary/src/xenia/kernel/kernel_state.cc#L1370)
+/// → [`Processor::Execute`](../../../../xenia-canary/src/xenia/cpu/processor.cc#L413)
+/// path: pick a guest thread, borrow its `PpcContext`, jam the ISR
+/// PC + args into it, and **run the interpreter inline on the host
+/// thread** until the ISR returns to `LR_HALT_SENTINEL`. Then restore
+/// the borrowed context and continue.
 ///
-/// Unlike the earlier P6 version which only delivered when HW 0 was
-/// `Ready`, this one also delivers when HW 0 is `Blocked`: the injector
-/// stashes the block reason into the new `HwState::ServicingIrq(reason)`
-/// variant, flips the thread to that state so `round_schedule` runs it,
-/// and — on callback return to `LR_HALT_SENTINEL` — the restore path
-/// re-creates `Blocked(reason)`, unless a `wake()` during the callback
-/// (e.g. `KeSetEvent` → `wake_eligible_waiters`) flipped it to `Ready`,
-/// in which case the wait was resolved and we leave it.
+/// Drains the full pending FIFO each call — canary's frame-limiter
+/// runs at its own cadence and our queue can already hold up to
+/// `INTERRUPT_QUEUE_CAP` coalesced v-sync events.
 ///
-/// This is the fix that unblocks games (like Sylpheed) which gate their
-/// main loop on a v-sync callback signaling an event the main thread
-/// waits on. The earlier "only-when-Ready" policy dropped 397 of 399
-/// observed v-syncs on a 1 B-instruction Sylpheed probe; now they
-/// actually get delivered.
-fn try_inject_graphics_interrupt(kernel: &mut xenia_kernel::KernelState) {
+/// Why this replaces the prior victim-mutate-then-wait scheme: with
+/// the old asynchronous injection, when every guest thread idled (post
+/// boot, when Sylpheed's main thread reaches its WAIT_FOREVER on the
+/// vsync-driven PKEVENT and all worker threads are likewise Blocked),
+/// the next scheduler round had no `Ready` victim and `Blocked` ones
+/// still required at least one round of execution to reach the
+/// callback. Audit-059 measured `gpu.interrupt.delivered = 54` over
+/// 3.9 s vs canary's 4712 — an 87× shortfall. Host-driven dispatch
+/// makes delivery rate a function of wall clock, not guest-thread
+/// readiness.
+///
+/// Victim selection still mirrors the canary precedent: prefer Ready
+/// (no state mangling), else any Blocked thread (we temporarily flip
+/// to `ServicingIrq(reason)` for the duration of the inline run so
+/// `call_export` etc. see a coherent thread state, and restore the
+/// `Blocked(reason)` on the way out unless the ISR itself signaled a
+/// wake). Idle / Exited / already-ServicingIrq slots are skipped — if
+/// nothing remains the source is dropped (still the right behavior;
+/// canary's `XThread::GetCurrentThread()` would assert).
+///
+/// All execution while in-flight runs against the borrowed thread's
+/// `ctx`. We set `scheduler.current = Some(target_ref)` so kernel
+/// imports (`KeSetEvent`, `KeReleaseSemaphore`, etc.) reach the right
+/// context, then restore the previous `current` on the way out. The
+/// dispatch is single-threaded — under `--parallel` it runs on the
+/// coordinator with workers parked at the phaser barrier, so there is
+/// no contention.
+fn dispatch_graphics_interrupts(
+    kernel: &mut xenia_kernel::KernelState,
+    mem: &xenia_memory::GuestMemory,
+    stats: &mut ExecStats,
+    decode_cache: &mut xenia_cpu::decoder::DecodeCache,
+    thunk_map: &HashMap<u32, (ModuleId, u16, String)>,
+) {
+    use xenia_cpu::interpreter::{step_cached, StepResult};
     use xenia_cpu::scheduler::HwState;
+    const LR_HALT: u32 = xenia_cpu::context::LR_HALT_SENTINEL as u32;
+    /// Defensive cap so a runaway ISR can't lock the coordinator on
+    /// the per-tick dispatch. Real Sylpheed vsync ISR is ~40 PPC
+    /// instructions; canary's `Processor::Execute` has no analogous
+    /// cap because it runs on a dedicated host thread, but we run
+    /// inline on the coordinator so a budget is prudent.
+    const MAX_INSTRS_PER_ISR: u64 = 1_000_000;
 
-    if kernel.interrupts.is_in_callback() {
-        return;
-    }
     let Some(cb) = kernel.interrupts.callback else {
-        // No callback registered; drain any pending entries (they
-        // wouldn't have made it into the queue per `queue_interrupt`'s
-        // own `callback.is_none()` guard, but be defensive).
         kernel.interrupts.pending.clear();
         return;
     };
 
-    let Some(source) = kernel.interrupts.peek_next() else {
-        return;
+    // Iterate-2.BF.γ: graphics dispatch is fully synchronous (host-driven,
+    // iterate-2.BE) — it borrows a guest thread, runs the ISR to
+    // LR_HALT_SENTINEL, and restores all in-call before returning. So it
+    // CAN safely coexist with an audio callback mid-flight, *as long as we
+    // pick a different victim thread* than the one audio borrowed. The old
+    // blanket `is_in_callback()` gate caused 5.85M skipped dispatches in
+    // lockstep boot (vs 55 with-pending dispatches) — audio is essentially
+    // always mid-flight on its dedicated worker, which choked vsync
+    // delivery at ~54. Exclude only audio's borrowed thread; the queue
+    // drains synchronously and graphics ISR completion does not touch
+    // `interrupts.saved` (used exclusively by the async audio path).
+    let audio_borrowed = if kernel.interrupts.is_in_callback() {
+        kernel.interrupts.injected_ref
+    } else {
+        None
     };
 
-    // Canary's `EmulateCPInterruptDPC` (kernel_state.cc:1373) dispatches on
-    // whatever the current thread happens to be — real hardware fires the
-    // interrupt on CPU 2 and the kernel impersonates a DPC on top of
-    // whichever thread is active. Hard-anchoring to HW 0 breaks the moment
-    // `main()` returns: Sylpheed's main thread exits right after init, the
-    // render worker spins on a `PKEVENT` inside the interrupt callback's
-    // user_data struct (`user_data + 0x5C`), and because HW 0 is now
-    // `Exited(_)` our injector drops every subsequent vsync — the PKEVENT
-    // is never signaled and the worker polls forever.
-    //
-    // Pick the first HW thread we can plausibly run the callback on:
-    //   1. Prefer `Ready` (no state-mangling needed)
-    //   2. Else take a `Blocked(reason)` thread and swap to
-    //      `ServicingIrq(reason)` so the round scheduler runs it; the
-    //      LR-sentinel restore path reinstates the block on callback return
-    //   3. Skip `Idle`, `Exited`, or already-`ServicingIrq` slots
-    //
-    // The callback itself just signals a game-side event and returns — it
-    // doesn't care which HW thread it ran on.
-    // Pass 1: find any Ready thread across all slots.
-    let mut victim: Option<xenia_cpu::ThreadRef> = None;
-    'outer_ready: for (hw_id, slot) in kernel.scheduler.slots.iter().enumerate() {
-        for (idx, t) in slot.runqueue.iter().enumerate() {
-            if matches!(t.state, HwState::Ready) {
-                victim = Some(xenia_cpu::ThreadRef::new(hw_id as u8, idx as u16));
-                break 'outer_ready;
-            }
-        }
-    }
-    // Pass 2: any Blocked thread (we'll flip it to ServicingIrq).
-    if victim.is_none() {
-        'outer_blocked: for (hw_id, slot) in kernel.scheduler.slots.iter().enumerate() {
+    while let Some(source) = kernel.interrupts.peek_next() {
+        // Victim selection: Ready first, then Blocked (canary's
+        // `XThread::GetCurrentThread()` analog — any live thread will
+        // do for borrowing context). Skip Idle/Exited/ServicingIrq.
+        // Skip the audio-borrowed thread (if any) to avoid clobbering
+        // its `SavedCallbackCtx` mid-flight.
+        let excluded = audio_borrowed;
+        let mut victim: Option<xenia_cpu::ThreadRef> = None;
+        'outer_ready: for (hw_id, slot) in kernel.scheduler.slots.iter().enumerate() {
             for (idx, t) in slot.runqueue.iter().enumerate() {
-                if matches!(t.state, HwState::Blocked(_)) {
-                    victim = Some(xenia_cpu::ThreadRef::new(hw_id as u8, idx as u16));
-                    break 'outer_blocked;
+                let r = xenia_cpu::ThreadRef::new(hw_id as u8, idx as u16);
+                if excluded == Some(r) {
+                    continue;
+                }
+                if matches!(t.state, HwState::Ready) {
+                    victim = Some(r);
+                    break 'outer_ready;
                 }
             }
         }
-    }
-    let Some(target_ref) = victim else {
-        // All threads Idle/Exited/already servicing — nothing to inject on.
-        kernel.interrupts.take_next();
-        kernel.interrupts.dropped += 1;
-        return;
-    };
-
-    let t = kernel.scheduler.thread_mut(target_ref);
-    let prev_state = t.state.clone();
-    match prev_state {
-        HwState::Ready => {}
-        HwState::Blocked(reason) => {
-            t.state = HwState::ServicingIrq(reason);
+        if victim.is_none() {
+            'outer_blocked: for (hw_id, slot) in kernel.scheduler.slots.iter().enumerate() {
+                for (idx, t) in slot.runqueue.iter().enumerate() {
+                    let r = xenia_cpu::ThreadRef::new(hw_id as u8, idx as u16);
+                    if excluded == Some(r) {
+                        continue;
+                    }
+                    if matches!(t.state, HwState::Blocked(_)) {
+                        victim = Some(r);
+                        break 'outer_blocked;
+                    }
+                }
+            }
         }
-        _ => unreachable!("victim selection above filtered out other variants"),
-    }
+        let Some(target_ref) = victim else {
+            // No donor at all — drop and exit (no point looping if the
+            // next source has the same problem).
+            kernel.interrupts.take_next();
+            kernel.interrupts.dropped += 1;
+            return;
+        };
 
-    let _ = kernel.interrupts.take_next();
-    let t = kernel.scheduler.thread_mut(target_ref);
-    let saved = xenia_kernel::SavedCallbackCtx::capture(&t.ctx, source);
-    kernel.interrupts.injected_ref = Some(target_ref);
-    t.ctx.pc = cb.callback_pc;
-    t.ctx.lr = xenia_cpu::context::LR_HALT_SENTINEL;
-    // Canary `Processor::Execute` decrements the guest SP by 176 before
-    // running the callback and restores on return (see Canary
-    // processor.cc:383). Without this pad the callback's
-    // `__savegprlr_N` prologue stomps the interrupted function's
-    // already-saved LR at [r1-8], so when the interrupted function
-    // later returns via `__restgprlr_N -> bclr` it jumps to
-    // `LR_HALT_SENTINEL` and the thread exits prematurely. Matching
-    // restore lives in `SavedCallbackCtx::restore` (which now also
-    // restores r1).
-    t.ctx.gpr[1] = t
-        .ctx
-        .gpr[1]
-        .wrapping_sub(xenia_kernel::interrupts::CALLBACK_STACK_PAD as u64);
-    t.ctx.gpr[3] = source as u64;
-    t.ctx.gpr[4] = cb.user_data as u64;
-    kernel.interrupts.saved = Some(saved);
-    metrics::counter!("gpu.interrupt.delivered", "source" => format!("{source}"))
-        .increment(1);
-    tracing::debug!(
-        source,
-        hw_id = target_ref.hw_id,
-        idx = target_ref.idx,
-        callback = format_args!("{:#010x}", cb.callback_pc),
-        "graphics interrupt: injecting"
-    );
+        // Commit: pop the queue, flag temporary state.
+        let _ = kernel.interrupts.take_next();
+        let prev_state = kernel.scheduler.thread(target_ref).state.clone();
+        let was_blocked = matches!(prev_state, HwState::Blocked(_));
+        if let HwState::Blocked(reason) = prev_state.clone() {
+            kernel.scheduler.thread_mut(target_ref).state =
+                HwState::ServicingIrq(reason);
+        }
+
+        // Save the borrowed ctx fields the ISR will clobber. Matches
+        // canary's processor.cc:387-394 (save prev lr, run, restore).
+        let saved = {
+            let t = kernel.scheduler.thread_mut(target_ref);
+            let saved = xenia_kernel::SavedCallbackCtx::capture(&t.ctx, source);
+            t.ctx.pc = cb.callback_pc;
+            t.ctx.lr = xenia_cpu::context::LR_HALT_SENTINEL;
+            // Canary processor.cc:383 — pad SP so the callback's
+            // __savegprlr_N prologue doesn't stomp the interrupted
+            // function's saved LR at [r1-8].
+            t.ctx.gpr[1] = t
+                .ctx
+                .gpr[1]
+                .wrapping_sub(xenia_kernel::interrupts::CALLBACK_STACK_PAD as u64);
+            t.ctx.gpr[3] = source as u64;
+            t.ctx.gpr[4] = cb.user_data as u64;
+            saved
+        };
+
+        // Stash the previous `scheduler.current` (call_export reaches
+        // it; imports the ISR calls must dispatch on the borrowed
+        // thread). Restore on the way out.
+        let prev_current = kernel.scheduler.current;
+        kernel.scheduler.current = Some(target_ref);
+
+        metrics::counter!("gpu.interrupt.delivered", "source" => format!("{source}"))
+            .increment(1);
+        tracing::debug!(
+            source,
+            hw_id = target_ref.hw_id,
+            idx = target_ref.idx,
+            callback = format_args!("{:#010x}", cb.callback_pc),
+            "graphics interrupt: dispatching synchronously (iterate-2.BE)"
+        );
+
+        // Inline interpreter loop on the borrowed context until the
+        // ISR returns to LR_HALT_SENTINEL (its `blr` writes
+        // `lr → pc`). Per-instruction step handles imports via
+        // thunk_map (the ISR typically just calls `KeSetEvent`).
+        let mut isr_instrs: u64 = 0;
+        loop {
+            let pc = kernel.scheduler.ctx_mut_ref(target_ref).pc;
+            if pc == LR_HALT {
+                break;
+            }
+            if isr_instrs >= MAX_INSTRS_PER_ISR {
+                tracing::warn!(
+                    pc = format_args!("{:#010x}", pc),
+                    isr_instrs,
+                    "graphics ISR exceeded MAX_INSTRS_PER_ISR; aborting"
+                );
+                break;
+            }
+
+            // Import-thunk intercept: same shape as worker_prologue's
+            // step 2 (line ~2287).
+            if let Some((module, ordinal, _name)) = thunk_map.get(&pc) {
+                let module = *module;
+                let ordinal_u32 = *ordinal as u32;
+                kernel.call_export(module, ordinal_u32, mem);
+                let post_ref = kernel.scheduler.current;
+                let c = match post_ref {
+                    Some(r) => kernel.scheduler.ctx_mut_ref(r),
+                    None => kernel.scheduler.ctx_mut_ref(target_ref),
+                };
+                c.pc = c.lr as u32;
+                c.cycle_count += 1;
+                c.timebase += 1;
+                stats.instruction_count += 1;
+                stats.import_count += 1;
+                isr_instrs += 1;
+                continue;
+            }
+
+            if !mem.is_mapped(pc) {
+                tracing::error!(
+                    pc = format_args!("{:#010x}", pc),
+                    isr_instrs,
+                    "graphics ISR hit unmapped PC; aborting"
+                );
+                break;
+            }
+
+            let ctx = kernel.scheduler.ctx_mut_ref(target_ref);
+            let page_ver = mem.page_version(ctx.pc);
+            let r = step_cached(ctx, mem, decode_cache, page_ver);
+            stats.instruction_count += 1;
+            isr_instrs += 1;
+            match r {
+                StepResult::Continue => {}
+                StepResult::SystemCall => {
+                    tracing::warn!("graphics ISR hit `sc` instruction; aborting");
+                    break;
+                }
+                StepResult::Trap => {
+                    tracing::warn!("graphics ISR hit trap; aborting");
+                    break;
+                }
+                StepResult::Halted => break,
+                StepResult::Unimplemented(op) => {
+                    tracing::warn!(?op, "graphics ISR hit unimplemented opcode; aborting");
+                    break;
+                }
+            }
+        }
+
+        // Restore the borrowed context.
+        saved.restore(kernel.scheduler.ctx_mut_ref(target_ref));
+        kernel.scheduler.current = prev_current;
+        kernel.interrupts.delivered += 1;
+
+        // Restore thread state. If the ISR signaled a wake on the
+        // borrowed thread (e.g. canary `KeSetEvent` → scheduler wake)
+        // the state may already be Ready; only re-block if still
+        // ServicingIrq.
+        if was_blocked {
+            let t = kernel.scheduler.thread_mut(target_ref);
+            if let HwState::ServicingIrq(reason) = t.state.clone() {
+                t.state = HwState::Blocked(reason);
+            }
+        }
+    }
 }
 
 /// AUDIT-032 Plan B — inject a pending XAudio buffer-complete callback
 /// into the **dedicated audio worker** registered for the head-of-queue
-/// client. Mirrors
-/// [`try_inject_graphics_interrupt`] (same SP-pad, same saved-context
-/// restore-on-sentinel) but the target thread is fixed at registration
-/// time instead of selected via the random-victim policy. The pre-fix
+/// client. Uses the asynchronous LR-sentinel injection mechanism (same
+/// SP-pad, same `SavedCallbackCtx` restore-on-sentinel as the pre-iterate-2.BE
+/// graphics path) but the target thread is fixed at registration time
+/// instead of selected via the random-victim policy. The pre-fix
 /// random-victim path corrupted unrelated thread state
 /// (APUBUG-PRODUCER-001 "HW-thread hijack"); per-client workers eliminate
 /// that whole class of regression.
 ///
-/// Mutual exclusion with the graphics path is via the shared
-/// `interrupts.saved` slot — if a graphics callback is already in flight,
-/// `is_in_callback()` returns true and we bail until it returns to the
-/// `LR_HALT_SENTINEL`.
+/// Mutual exclusion with the graphics path (which is now synchronous —
+/// see `dispatch_graphics_interrupts`) is via the shared
+/// `interrupts.saved` slot — if an audio callback is already in flight,
+/// `is_in_callback()` returns true and `dispatch_graphics_interrupts`
+/// defers until it returns to the `LR_HALT_SENTINEL`.
 fn try_inject_audio_callback(kernel: &mut xenia_kernel::KernelState) {
     use xenia_cpu::scheduler::HwState;
 

crates/xenia-kernel/src/interrupts.rs

@@ -8,13 +8,18 @@
 //!    guest-issued command stream; source code 1 (`INTERRUPT_SOURCE_CP`).
 //!
 //! Canary's [xboxkrnl_video.cc:303-310](xenia-canary/src/xenia/kernel/xboxkrnl/xboxkrnl_video.cc#L303-L310)
-//! dispatches the callback on HW thread 0. We follow the same convention.
+//! dispatches the callback on HW thread 0. We follow the same convention
+//! for picking a *context donor*, but as of iterate-2.BE the dispatch
+//! itself is **synchronous and host-driven**: the main loop runs the ISR
+//! inline on the borrowed guest context, mirroring canary's
+//! `EmulateCPInterruptDPC → Processor::Execute` path
+//! ([kernel_state.cc:1370](../../../../xenia-canary/src/xenia/kernel/kernel_state.cc#L1370),
+//! [processor.cc:413](../../../../xenia-canary/src/xenia/cpu/processor.cc#L413)).
+//! Independent of whether the donor guest thread was Ready or Blocked.
 //!
-//! The delivery model is cooperative: we inject the callback entry into HW
-//! thread 0 at the top of a scheduler round when it's safe (not mid-export,
-//! not already inside another interrupt). When the callback returns to
-//! [`LR_HALT_SENTINEL`] the main loop restores the saved [`PpcContext`]
-//! fields and the HW thread picks up where it left off.
+//! The audio callback path (audit-048) still uses asynchronous LR-sentinel
+//! injection on a dedicated per-client worker thread; the
+//! [`SavedCallbackCtx`] machinery below remains in use there.
 
 use std::collections::VecDeque;
 use std::time::{Duration, Instant};

crates/xenia-kernel/src/state.rs

@@ -244,6 +244,41 @@ pub struct KernelState {
     /// Distinct from `ctor_probe_pcs` because that helper emits 8
     /// frames of back-chain per hit — too noisy for branch tracing.
     pub branch_probe_pcs: std::collections::HashSet<u32>,
+    /// AUDIT-2BF — diagnostic. PCs at which to emit a structured one-line
+    /// `AUDIT-PC-PROBE` record on every fire, designed for the silph init
+    /// chain virtual-dispatch site at `sub_82172BA0+0x1E8` (PC
+    /// `0x82172D88`, a `bctrl` after a 3-deep load of vtable slot 6). The
+    /// emitted line carries (pc, tid, hw, cycle, lr, r3, r11) plus four
+    /// guest-memory dereferences off `r3`: `[r3+0]` (vtable), `[[r3+0]+24]`
+    /// (slot 6 method pointer = the bctrl target), `[r3+0x0C]` (audit-059
+    /// round-9 canary-known auxiliary handle `0xF80000D8`), and `[r3+0x30]`
+    /// (canary-known embedded sub-object vtable `0x820A1870`). Distinct
+    /// from `branch_probe_pcs` because that helper only logs registers (no
+    /// memory) and from `lr_trace_pcs` because that emits JSON intended
+    /// for canary diffing, not the four hard-coded indirect dereferences
+    /// needed here. Read-only — no guest state mutation. Lockstep
+    /// digest unaffected. Settable via `--audit-pc-probe-hex` /
+    /// `XENIA_AUDIT_PC_PROBE`.
+    pub audit_pc_probe_pcs: std::collections::HashSet<u32>,
+    /// AUDIT-2BF round 14 — diagnostic. Optional guest VA. When set, each
+    /// `AUDIT-PC-PROBE` fire emits a paired `AUDIT-MEM-READ` line with
+    /// `addr`, `*addr` (singleton value), `**addr` (vtable), `***addr+0`
+    /// (vtable[0] = first virtual method), and `***addr+24` (vtable[6]
+    /// in 4-byte stride = slot 6 = silph chain bctrl target). Three-deep
+    /// dereference to resolve the vtable[0] target at the bctrl site
+    /// `0x822F1B4C` inside `sub_822F1AA8`. Read-only; lockstep digest
+    /// unaffected. Settable via `--audit-mem-read-hex` /
+    /// `XENIA_AUDIT_MEM_READ`.
+    pub audit_mem_read_addr: Option<u32>,
+    /// AUDIT-052 — diagnostic. When set, each `AUDIT-PC-PROBE` fire
+    /// additionally emits an `AUDIT-R3-DUMP` line with N bytes of guest
+    /// memory dumped from `r3` as `u32` lanes (4-byte aligned only).
+    /// Sized for audit-051's 80-byte stack-local struct at `r31+96`
+    /// inside `sub_82452DC0` (probe `sub_8245B000` entry where
+    /// `r3 == parent's r31+96`). Read-only; lockstep digest unaffected.
+    /// Settable via `--audit-r3-dump-bytes` /
+    /// `XENIA_AUDIT_R3_DUMP_BYTES`.
+    pub audit_r3_dump_bytes: Option<u32>,
     /// M12 — diagnostic. PCs at which to emit a structured JSONL record
     /// per fire, designed for diffing against xenia-canary's
     /// `--log_lr_on_pc` patch output. Each line carries
@@ -327,6 +362,9 @@ impl KernelState {
             ctor_probe_pcs: std::collections::HashSet::new(),
             pc_probe_consumers: HashMap::new(),
             branch_probe_pcs: std::collections::HashSet::new(),
+            audit_pc_probe_pcs: std::collections::HashSet::new(),
+            audit_mem_read_addr: None,
+            audit_r3_dump_bytes: None,
             lr_trace_pcs: std::collections::HashSet::new(),
             lr_trace_writer: None,
             dump_addrs: Vec::new(),
@@ -797,6 +835,91 @@ impl KernelState {
         );
     }
 
+    /// AUDIT-2BF — diagnostic. If the live PC for HW slot `hw_id` is in
+    /// `self.audit_pc_probe_pcs`, emit a single one-line
+    /// `AUDIT-PC-PROBE` record with (pc, tid, hw, cycle, lr, r3, r11)
+    /// plus four guest-memory dereferences off r3: `[r3+0]` (vtable),
+    /// `[[r3+0]+24]` (slot 6 method = bctrl target), `[r3+0x0C]`
+    /// (auxiliary handle field), `[r3+0x30]` (embedded sub-object
+    /// vtable field). Tuned for the silph init chain virtual-dispatch
+    /// site at `sub_82172BA0+0x1E8` (PC `0x82172D88`).
+    ///
+    /// Read-only. No guest-state mutation; lockstep digest unaffected.
+    /// Empty set is the common case → single `is_empty()` test on the
+    /// hot path.
+    pub fn fire_audit_pc_probe_if_match(&self, hw_id: u8, mem: &GuestMemory) {
+        if self.audit_pc_probe_pcs.is_empty() {
+            return;
+        }
+        let ctx = self.scheduler.ctx(hw_id);
+        let pc = ctx.pc;
+        if !self.audit_pc_probe_pcs.contains(&pc) {
+            return;
+        }
+        let tid = self.scheduler.tid(hw_id).unwrap_or(0);
+        let r3 = ctx.gpr[3] as u32;
+        let r11 = ctx.gpr[11] as u32;
+        let lr = ctx.lr as u32;
+        let cycle = ctx.cycle_count;
+        // Memory dereferences. Guest pointers may be unmapped/garbage;
+        // `read_u32` returns 0 for unmapped pages (heap.rs:510 returns
+        // a default), so an all-zero block in the output reliably
+        // indicates an invalid `r3`.
+        let vtable = mem.read_u32(r3);
+        let slot6_method = if vtable != 0 {
+            mem.read_u32(vtable.wrapping_add(24))
+        } else {
+            0
+        };
+        let aux_handle = mem.read_u32(r3.wrapping_add(0x0C));
+        let sub_vt = mem.read_u32(r3.wrapping_add(0x30));
+        println!(
+            "AUDIT-PC-PROBE pc={:#010x} tid={} hw={} cycle={} lr={:#010x} r3={:#010x} r11={:#010x} \
+             [r3+0]={:#010x} [[r3+0]+24]={:#010x} [r3+0x0C]={:#010x} [r3+0x30]={:#010x}",
+            pc, tid, hw_id, cycle, lr, r3, r11,
+            vtable, slot6_method, aux_handle, sub_vt,
+        );
+        // AUDIT-2BF round 14 — paired memory-read. When
+        // `audit_mem_read_addr` is set, dereference 3 deep: singleton
+        // pointer → vtable → vtable[0] / vtable[24]. Defensively
+        // null-checks each level. `read_u32` returns 0 for unmapped
+        // pages so all-zero output is the unmapped/uninitialized
+        // signature.
+        if let Some(addr) = self.audit_mem_read_addr {
+            let val = mem.read_u32(addr);
+            let vt = if val != 0 { mem.read_u32(val) } else { 0 };
+            let m0 = if vt != 0 { mem.read_u32(vt) } else { 0 };
+            let m6 = if vt != 0 { mem.read_u32(vt.wrapping_add(24)) } else { 0 };
+            println!(
+                "AUDIT-MEM-READ addr={:#010x} val={:#010x} vtable={:#010x} \
+                 vtable[0]={:#010x} vtable[24]={:#010x} pc={:#010x} tid={} cycle={}",
+                addr, val, vt, m0, m6, pc, tid, cycle,
+            );
+        }
+        // AUDIT-052 — dump N bytes of guest memory from r3 as u32 lanes
+        // when `audit_r3_dump_bytes` is set. Sized for the 80-byte
+        // stack-local struct at sub_82452DC0's `r31+96` (probe is
+        // sub_8245B000 entry where r3 IS the struct ptr). Output
+        // format: `AUDIT-R3-DUMP pc=… r3=… +0x00=… +0x04=… …`.
+        if let Some(n) = self.audit_r3_dump_bytes {
+            let n = n.min(256) & !3u32; // cap 256B, 4-byte align
+            let mut out = String::with_capacity(64 + (n as usize) * 16);
+            use std::fmt::Write as _;
+            let _ = write!(
+                &mut out,
+                "AUDIT-R3-DUMP pc={:#010x} tid={} cycle={} r3={:#010x}",
+                pc, tid, cycle, r3,
+            );
+            let mut off: u32 = 0;
+            while off < n {
+                let v = mem.read_u32(r3.wrapping_add(off));
+                let _ = write!(&mut out, " +0x{:02x}={:#010x}", off, v);
+                off = off.wrapping_add(4);
+            }
+            println!("{}", out);
+        }
+    }
+
     /// M12 — diagnostic. If the live PC for HW slot `hw_id` is in
     /// `self.lr_trace_pcs`, emit one JSONL record. Format mirrors what
     /// xenia-canary's `--log_lr_on_pc` patch emits, plus the cycle