Files
xenia-rs/audit-runs/audit-067-vptr-install-mem-watch/canary-patch-audit-67.diff
MechaCat02 ef93a4fa14 handoff: VSync/event-wedge fixes + iterate 2.A–2.BC research notes
Source changes (dormant parity infra, retained from iterate 2.AI/2.AO):
- xenia-kernel/exports.rs: nt_create_event manual_reset polarity +
  related event wiring
- xenia-gpu/mmio_region.rs: D1MODE_VBLANK_VLINE_STATUS hardcode parity

Also lands the audit-runs/ analysis notes (.md/.txt/.json digests) for the
iterate 2.x VSync/0x10e8/0x1004 wedge investigation. Raw trace dumps
(.jsonl/.gz/.csv/.stdout) and agent worktrees (.claude/) are gitignored as
regenerable local artifacts — see memory + HANDOFF for the running findings.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-05 07:19:08 +02:00

666 lines
25 KiB
Diff

diff --git a/src/xenia/cpu/backend/x64/x64_emitter.cc b/src/xenia/cpu/backend/x64/x64_emitter.cc
index 5da8f6adc..cbac9826c 100644
--- a/src/xenia/cpu/backend/x64/x64_emitter.cc
+++ b/src/xenia/cpu/backend/x64/x64_emitter.cc
@@ -13,6 +13,8 @@
#include <climits>
#include <cstring>
+#include <string>
+#include <vector>
#include "third_party/fmt/include/fmt/format.h"
#include "xenia/base/assert.h"
@@ -63,6 +65,76 @@ DEFINE_bool(instrument_call_times, false,
"Compute time taken for functions, for profiling guest code",
"x64");
#endif
+
+// AUDIT-061/067: forward decls of probe/watch tables (defined in
+// ppc_hir_builder.cc).
+namespace xe {
+namespace cpu {
+namespace audit61 {
+const std::vector<uint32_t>& pcs();
+} // namespace audit61
+namespace audit67 {
+const std::vector<uint32_t>& vals();
+} // namespace audit67
+} // namespace cpu
+} // namespace xe
+
+// AUDIT-061: handler for trap codes [200, 232). arg0 carries trap idx
+// (trap_code - 200), mapping to ::xe::cpu::audit61::pcs()[idx]. Emits one
+// log line per fire with cr0/cr6 LGE flags + key GPRs + LR + tid.
+static uint64_t TrapAudit61Branch(void* raw_context, uint64_t idx) {
+ auto* ctx = reinterpret_cast<xe::cpu::ppc::PPCContext_s*>(raw_context);
+ const auto& pcs = ::xe::cpu::audit61::pcs();
+ uint32_t pc = (idx < pcs.size()) ? pcs[static_cast<size_t>(idx)] : 0u;
+ uint32_t tid = 0;
+ if (ctx->thread_state) {
+ tid = ctx->thread_state->thread_id();
+ }
+ auto enc = [](uint8_t lt, uint8_t gt, uint8_t eq) {
+ char buf[4];
+ buf[0] = lt ? 'L' : '.';
+ buf[1] = gt ? 'G' : '.';
+ buf[2] = eq ? 'E' : '.';
+ buf[3] = '\0';
+ return std::string(buf);
+ };
+ XELOGI(
+ "AUDIT-061-BR pc={:08X} lr={:08X} cr0={} cr6={} r3={:08X} r4={:08X} "
+ "r5={:08X} r6={:08X} r31={:08X} tid={}",
+ pc, static_cast<uint32_t>(ctx->lr),
+ enc(ctx->cr0.cr0_lt, ctx->cr0.cr0_gt, ctx->cr0.cr0_eq),
+ enc(ctx->cr6.cr6_all_equal, ctx->cr6.cr6_1, ctx->cr6.cr6_none_equal),
+ static_cast<uint32_t>(ctx->r[3]), static_cast<uint32_t>(ctx->r[4]),
+ static_cast<uint32_t>(ctx->r[5]), static_cast<uint32_t>(ctx->r[6]),
+ static_cast<uint32_t>(ctx->r[31]), tid);
+ return 0;
+}
+
+// AUDIT-067: handler for trap codes [250, 254). arg0 carries trap idx
+// (trap_code - 250), mapping to ::xe::cpu::audit67::vals()[idx]. Fired when
+// a 4-byte guest store sees the configured value. The store-emit site stashed
+// (pc << 32) | (ea & 0xFFFFFFFF) into ctx->scratch right before the trap.
+static uint64_t TrapAudit67ValueWatch(void* raw_context, uint64_t idx) {
+ auto* ctx = reinterpret_cast<xe::cpu::ppc::PPCContext_s*>(raw_context);
+ const auto& vals = ::xe::cpu::audit67::vals();
+ uint32_t val =
+ (idx < vals.size()) ? vals[static_cast<size_t>(idx)] : 0u;
+ uint32_t pc = static_cast<uint32_t>(ctx->scratch >> 32);
+ uint32_t dst = static_cast<uint32_t>(ctx->scratch & 0xFFFFFFFFu);
+ uint32_t tid = 0;
+ if (ctx->thread_state) {
+ tid = ctx->thread_state->thread_id();
+ }
+ XELOGI(
+ "AUDIT-067-VAL pc={:08X} lr={:08X} val={:08X} dst={:08X} "
+ "r3={:08X} r4={:08X} r5={:08X} r6={:08X} r31={:08X} tid={}",
+ pc, static_cast<uint32_t>(ctx->lr), val, dst,
+ static_cast<uint32_t>(ctx->r[3]), static_cast<uint32_t>(ctx->r[4]),
+ static_cast<uint32_t>(ctx->r[5]), static_cast<uint32_t>(ctx->r[6]),
+ static_cast<uint32_t>(ctx->r[31]), tid);
+ return 0;
+}
+
namespace xe {
namespace cpu {
namespace backend {
@@ -455,6 +527,20 @@ void X64Emitter::Trap(uint16_t trap_type) {
// ?
break;
default:
+ // AUDIT-067: trap codes [250, 254) dispatch the value-watch handler.
+ // arg0 = idx into ::xe::cpu::audit67::vals().
+ if (trap_type >= 250 && trap_type < 254) {
+ CallNative(::TrapAudit67ValueWatch,
+ static_cast<uint64_t>(trap_type - 250));
+ break;
+ }
+ // AUDIT-061: trap codes [200, 232) dispatch the branch-probe handler.
+ // arg0 = idx into ::xe::cpu::audit61::pcs().
+ if (trap_type >= 200 && trap_type < 232) {
+ CallNative(::TrapAudit61Branch,
+ static_cast<uint64_t>(trap_type - 200));
+ break;
+ }
XELOGW("Unknown trap type {}", trap_type);
db(0xCC);
break;
diff --git a/src/xenia/cpu/cpu_flags.cc b/src/xenia/cpu/cpu_flags.cc
index 3ff067e15..f78dad157 100644
--- a/src/xenia/cpu/cpu_flags.cc
+++ b/src/xenia/cpu/cpu_flags.cc
@@ -57,3 +57,25 @@ DEFINE_bool(break_condition_truncate, true, "truncate value to 32-bits", "CPU");
DEFINE_bool(break_on_debugbreak, true, "int3 on JITed __debugbreak requests.",
"CPU");
+
+// AUDIT-DEMO: smoke marker (memory entry: emulator.cc:225,283). Always-on bool.
+DEFINE_bool(audit_demo_setup_trace, true,
+ "Audit smoke marker: log AUDIT-DEMO-SETUP-BEGIN at emulator setup.",
+ "Audit");
+
+// AUDIT-061: comma-separated list of guest PCs to log on each fire.
+// Format: "0xPC1,0xPC2,..." (max 32 PCs). Each fire emits
+// AUDIT-061-BR pc=X lr=X cr0=LGE cr6=LGE r3=X r4=X r5=X r6=X r31=X tid=N.
+// Default empty (off); no perf cost when empty.
+DEFINE_string(audit_61_branch_probe_pcs, "",
+ "AUDIT-061: CSV of guest PCs to trace (cr0/cr6 + regs/tid).",
+ "Audit");
+
+// AUDIT-067: comma-separated list of u32 values to watch. When non-empty,
+// every 4-byte guest store (stw/stwu/stwx/stwux/stmw) emits a runtime
+// equality check; matches log AUDIT-067-VAL pc=X lr=X val=X dst=X r3..r6 r31 tid=N.
+// Max 4 values. Default empty (off); zero overhead when empty.
+DEFINE_string(audit_67_value_watch, "",
+ "AUDIT-067: CSV of u32 values (max 4) — log every guest "
+ "store whose value matches.",
+ "Audit");
diff --git a/src/xenia/cpu/cpu_flags.h b/src/xenia/cpu/cpu_flags.h
index 38c4f98ba..5f52647b5 100644
--- a/src/xenia/cpu/cpu_flags.h
+++ b/src/xenia/cpu/cpu_flags.h
@@ -35,4 +35,16 @@ DECLARE_bool(break_condition_truncate);
DECLARE_bool(break_on_debugbreak);
+// AUDIT-DEMO smoke marker.
+DECLARE_bool(audit_demo_setup_trace);
+
+// AUDIT-061: multi-PC branch probe — emits one log line per fire with
+// (pc, lr, cr0 LGE, cr6 LGE, r3, r4, r5, r6, r31, tid). CSV of guest PCs.
+DECLARE_string(audit_61_branch_probe_pcs);
+
+// AUDIT-067: value-watch — emit a log line for each 32-bit guest store whose
+// value-to-be-stored matches any configured value. CSV of u32 values
+// ("0xDEADBEEF,..."), max 4 entries. Default empty (off); zero cost when empty.
+DECLARE_string(audit_67_value_watch);
+
#endif // XENIA_CPU_CPU_FLAGS_H_
diff --git a/src/xenia/cpu/ppc/ppc_emit_altivec.cc b/src/xenia/cpu/ppc/ppc_emit_altivec.cc
index 513b21391..c9af025ff 100644
--- a/src/xenia/cpu/ppc/ppc_emit_altivec.cc
+++ b/src/xenia/cpu/ppc/ppc_emit_altivec.cc
@@ -9,12 +9,28 @@
#include "xenia/cpu/ppc/ppc_emit-private.h"
+#include <vector>
#include "xenia/base/assert.h"
+#include "xenia/cpu/cpu_flags.h"
#include "xenia/cpu/ppc/ppc_context.h"
#include "xenia/cpu/ppc/ppc_hir_builder.h"
#include <cmath>
+// AUDIT-067: forward-decls. Defined in ppc_emit_memory.cc / ppc_hir_builder.cc.
+namespace xe {
+namespace cpu {
+namespace audit67 {
+const std::vector<uint32_t>& vals();
+}
+namespace ppc {
+void EmitAudit67ValueWatchVec(PPCHIRBuilder& f, uint32_t pc,
+ ::xe::cpu::hir::Value* vec128,
+ ::xe::cpu::hir::Value* ea);
+}
+}
+}
+
namespace xe {
namespace cpu {
namespace ppc {
@@ -175,6 +191,21 @@ int InstrEmit_stvewx_(PPCHIRBuilder& f, const InstrData& i, uint32_t vd,
f.Shr(f.And(f.Truncate(ea, INT8_TYPE), f.LoadConstantUint8(0xF)), 2);
Value* v = f.Extract(f.LoadVR(vd), el, INT32_TYPE);
f.Store(ea, f.ByteSwap(v));
+ if (!::xe::cpu::audit67::vals().empty()) {
+ // For stvewx: only one lane is actually stored; piggyback on the scalar
+ // value-watch helper by emitting the equivalent of stw of v at ea.
+ Value* pc_hi64 =
+ f.LoadConstantUint64(static_cast<uint64_t>(i.address) << 32);
+ Value* ea_lo64 = f.ZeroExtend(f.Truncate(ea, INT32_TYPE), INT64_TYPE);
+ Value* packed = f.Or(pc_hi64, ea_lo64);
+ const auto& vals = ::xe::cpu::audit67::vals();
+ for (size_t idx = 0; idx < vals.size(); ++idx) {
+ Value* cmp = f.CompareEQ(v, f.LoadConstantUint32(vals[idx]));
+ f.StoreContext(offsetof(::xe::cpu::ppc::PPCContext, scratch), packed);
+ f.ContextBarrier();
+ f.TrapTrue(cmp, static_cast<uint16_t>(250 + idx));
+ }
+ }
return 0;
}
int InstrEmit_stvewx(PPCHIRBuilder& f, const InstrData& i) {
@@ -187,7 +218,11 @@ int InstrEmit_stvewx128(PPCHIRBuilder& f, const InstrData& i) {
int InstrEmit_stvx_(PPCHIRBuilder& f, const InstrData& i, uint32_t vd,
uint32_t ra, uint32_t rb) {
Value* ea = f.And(CalculateEA_0(f, ra, rb), f.LoadConstantUint64(~0xFull));
- f.Store(ea, f.ByteSwap(f.LoadVR(vd)));
+ Value* vec = f.LoadVR(vd);
+ f.Store(ea, f.ByteSwap(vec));
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatchVec(f, i.address, vec, ea);
+ }
return 0;
}
int InstrEmit_stvx(PPCHIRBuilder& f, const InstrData& i) {
diff --git a/src/xenia/cpu/ppc/ppc_emit_memory.cc b/src/xenia/cpu/ppc/ppc_emit_memory.cc
index b4bdabb49..a6b44697d 100644
--- a/src/xenia/cpu/ppc/ppc_emit_memory.cc
+++ b/src/xenia/cpu/ppc/ppc_emit_memory.cc
@@ -10,11 +10,22 @@
#include "xenia/cpu/ppc/ppc_emit-private.h"
#include <stddef.h>
+#include <vector>
#include "xenia/base/assert.h"
#include "xenia/base/cvar.h"
+#include "xenia/cpu/cpu_flags.h"
#include "xenia/cpu/ppc/ppc_context.h"
#include "xenia/cpu/ppc/ppc_hir_builder.h"
+// AUDIT-067: forward-decl of value-watch table (defined in ppc_hir_builder.cc).
+namespace xe {
+namespace cpu {
+namespace audit67 {
+const std::vector<uint32_t>& vals();
+} // namespace audit67
+} // namespace cpu
+} // namespace xe
+
DEFINE_bool(
disable_prefetch_and_cachecontrol, true,
"Disables translating ppc prefetch/cache flush instructions to host "
@@ -67,6 +78,90 @@ void StoreEA(PPCHIRBuilder& f, uint32_t rt, Value* ea) {
f.StoreGPR(rt, ea);
}
+// AUDIT-067: emit a runtime equality check on the 32-bit value-to-be-stored
+// against each configured watch value. On match, store (pc, EA) packed into
+// the PPCContext scratch field so the native trap handler can read them,
+// then fire a trap with code (kTrapBase + idx). Done host-side as a
+// build-time pc constant + a runtime EA truncate, packed as
+// (pc << 32) | (ea & 0xFFFFFFFF) so the handler can decompose.
+static void EmitAudit67ValueWatch(PPCHIRBuilder& f, uint32_t pc, Value* val32,
+ Value* ea) {
+ const auto& vals = ::xe::cpu::audit67::vals();
+ if (vals.empty()) return;
+ // pc is known at JIT time → emit as constant; ea is runtime.
+ Value* pc_hi64 = f.LoadConstantUint64(static_cast<uint64_t>(pc) << 32);
+ Value* ea_lo64 = f.ZeroExtend(f.Truncate(ea, INT32_TYPE), INT64_TYPE);
+ Value* packed = f.Or(pc_hi64, ea_lo64);
+ for (size_t idx = 0; idx < vals.size(); ++idx) {
+ Value* cmp = f.CompareEQ(val32, f.LoadConstantUint32(vals[idx]));
+ f.StoreContext(offsetof(::xe::cpu::ppc::PPCContext, scratch), packed);
+ f.ContextBarrier();
+ f.TrapTrue(cmp, static_cast<uint16_t>(250 + idx));
+ }
+}
+
+// AUDIT-067 128-bit (vector) variant: checks each of the 4 32-bit lanes in a
+// vector store. Used for stvx/stvxl/stvewx (memcpy-derived installs may use
+// 128-bit vector stores). The matched lane is reflected in the dst by
+// adding (lane * 4) so the handler can see exactly where in memory the
+// value lands. Declared with external linkage so altivec.cc can call it.
+void EmitAudit67ValueWatchVec(PPCHIRBuilder& f, uint32_t pc,
+ Value* vec128, Value* ea) {
+ const auto& vals = ::xe::cpu::audit67::vals();
+ if (vals.empty()) return;
+ Value* pc_hi64 = f.LoadConstantUint64(static_cast<uint64_t>(pc) << 32);
+ for (size_t idx = 0; idx < vals.size(); ++idx) {
+ Value* watch = f.LoadConstantUint32(vals[idx]);
+ for (uint8_t lane = 0; lane < 4; ++lane) {
+ Value* lane_val = f.Extract(vec128, lane, INT32_TYPE);
+ Value* cmp = f.CompareEQ(lane_val, watch);
+ Value* lane_off = f.LoadConstantUint32(static_cast<uint32_t>(lane * 4));
+ Value* dst32 = f.Add(f.Truncate(ea, INT32_TYPE), lane_off);
+ Value* packed = f.Or(pc_hi64, f.ZeroExtend(dst32, INT64_TYPE));
+ f.StoreContext(offsetof(::xe::cpu::ppc::PPCContext, scratch), packed);
+ f.ContextBarrier();
+ f.TrapTrue(cmp, static_cast<uint16_t>(250 + idx));
+ }
+ }
+}
+
+// AUDIT-067 64-bit variant: same as above but checks BOTH halves of a 64-bit
+// stored value. EA points at the start of the 8-byte store; the matched half
+// is encoded into the trap idx via (250 + 2*idx + half), where half=0 means
+// upper 32 bits (lower address), half=1 means lower 32 bits (upper address).
+static void EmitAudit67ValueWatch64(PPCHIRBuilder& f, uint32_t pc, Value* val64,
+ Value* ea) {
+ const auto& vals = ::xe::cpu::audit67::vals();
+ if (vals.empty()) return;
+ // PowerPC is big-endian: u64 stored at EA places upper-32 bits at EA+0
+ // and lower-32 bits at EA+4. Check both halves against each watch value.
+ Value* upper32 = f.Truncate(f.Shr(val64, int8_t(32)), INT32_TYPE); // bits[63:32]
+ Value* lower32 = f.Truncate(val64, INT32_TYPE); // bits[31:0]
+ Value* pc_hi64 = f.LoadConstantUint64(static_cast<uint64_t>(pc) << 32);
+ for (size_t idx = 0; idx < vals.size(); ++idx) {
+ // Upper half lands at EA+0.
+ {
+ Value* cmp = f.CompareEQ(upper32, f.LoadConstantUint32(vals[idx]));
+ Value* ea_lo64 = f.ZeroExtend(f.Truncate(ea, INT32_TYPE), INT64_TYPE);
+ Value* packed = f.Or(pc_hi64, ea_lo64);
+ f.StoreContext(offsetof(::xe::cpu::ppc::PPCContext, scratch), packed);
+ f.ContextBarrier();
+ f.TrapTrue(cmp, static_cast<uint16_t>(250 + idx));
+ }
+ // Lower half lands at EA+4.
+ {
+ Value* cmp = f.CompareEQ(lower32, f.LoadConstantUint32(vals[idx]));
+ Value* ea_plus4 =
+ f.Add(f.Truncate(ea, INT32_TYPE), f.LoadConstantUint32(4));
+ Value* ea_lo64 = f.ZeroExtend(ea_plus4, INT64_TYPE);
+ Value* packed = f.Or(pc_hi64, ea_lo64);
+ f.StoreContext(offsetof(::xe::cpu::ppc::PPCContext, scratch), packed);
+ f.ContextBarrier();
+ f.TrapTrue(cmp, static_cast<uint16_t>(250 + idx));
+ }
+ }
+}
+
// Integer load (A-13)
int InstrEmit_lbz(PPCHIRBuilder& f, const InstrData& i) {
@@ -518,9 +613,11 @@ int InstrEmit_stw(PPCHIRBuilder& f, const InstrData& i) {
b = f.LoadGPR(i.D.RA);
}
Value* offset = f.LoadConstantInt64(XEEXTS16(i.D.DS));
- f.StoreOffset(b, offset,
- f.ByteSwap(f.Truncate(f.LoadGPR(i.D.RT), INT32_TYPE)));
-
+ Value* val32 = f.Truncate(f.LoadGPR(i.D.RT), INT32_TYPE);
+ f.StoreOffset(b, offset, f.ByteSwap(val32));
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch(f, i.address, val32, f.Add(b, offset));
+ }
return 0;
}
@@ -532,10 +629,14 @@ int InstrEmit_stmw(PPCHIRBuilder& f, const InstrData& i) {
b = f.LoadGPR(i.D.RA);
}
+ const bool watch_active = !::xe::cpu::audit67::vals().empty();
for (uint32_t j = 0; j < 32 - i.D.RT; ++j) {
Value* offset = f.LoadConstantInt64(XEEXTS16(i.D.DS) + j * 4);
- f.StoreOffset(b, offset,
- f.ByteSwap(f.Truncate(f.LoadGPR(i.D.RT + j), INT32_TYPE)));
+ Value* val32 = f.Truncate(f.LoadGPR(i.D.RT + j), INT32_TYPE);
+ f.StoreOffset(b, offset, f.ByteSwap(val32));
+ if (watch_active) {
+ EmitAudit67ValueWatch(f, i.address, val32, f.Add(b, offset));
+ }
}
return 0;
}
@@ -545,8 +646,12 @@ int InstrEmit_stwu(PPCHIRBuilder& f, const InstrData& i) {
// MEM(EA, 4) <- (RS)[32:63]
// RA <- EA
Value* ea = CalculateEA_i(f, i.D.RA, XEEXTS16(i.D.DS));
- f.Store(ea, f.ByteSwap(f.Truncate(f.LoadGPR(i.D.RT), INT32_TYPE)));
+ Value* val32 = f.Truncate(f.LoadGPR(i.D.RT), INT32_TYPE);
+ f.Store(ea, f.ByteSwap(val32));
StoreEA(f, i.D.RA, ea);
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch(f, i.address, val32, ea);
+ }
return 0;
}
@@ -555,8 +660,12 @@ int InstrEmit_stwux(PPCHIRBuilder& f, const InstrData& i) {
// MEM(EA, 4) <- (RS)[32:63]
// RA <- EA
Value* ea = CalculateEA(f, i.X.RA, i.X.RB);
- f.Store(ea, f.ByteSwap(f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE)));
+ Value* val32 = f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE);
+ f.Store(ea, f.ByteSwap(val32));
StoreEA(f, i.X.RA, ea);
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch(f, i.address, val32, ea);
+ }
return 0;
}
@@ -568,7 +677,11 @@ int InstrEmit_stwx(PPCHIRBuilder& f, const InstrData& i) {
// EA <- b + (RB)
// MEM(EA, 4) <- (RS)[32:63]
Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
- f.Store(ea, f.ByteSwap(f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE)));
+ Value* val32 = f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE);
+ f.Store(ea, f.ByteSwap(val32));
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch(f, i.address, val32, ea);
+ }
return 0;
}
@@ -587,7 +700,11 @@ int InstrEmit_std(PPCHIRBuilder& f, const InstrData& i) {
}
Value* offset = f.LoadConstantInt64(XEEXTS16(i.DS.DS << 2));
- f.StoreOffset(b, offset, f.ByteSwap(f.LoadGPR(i.DS.RT)));
+ Value* val64 = f.LoadGPR(i.DS.RT);
+ f.StoreOffset(b, offset, f.ByteSwap(val64));
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch64(f, i.address, val64, f.Add(b, offset));
+ }
return 0;
}
@@ -596,8 +713,12 @@ int InstrEmit_stdu(PPCHIRBuilder& f, const InstrData& i) {
// MEM(EA, 8) <- (RS)
// RA <- EA
Value* ea = CalculateEA_i(f, i.DS.RA, XEEXTS16(i.DS.DS << 2));
- f.Store(ea, f.ByteSwap(f.LoadGPR(i.DS.RT)));
+ Value* val64 = f.LoadGPR(i.DS.RT);
+ f.Store(ea, f.ByteSwap(val64));
StoreEA(f, i.DS.RA, ea);
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch64(f, i.address, val64, ea);
+ }
return 0;
}
@@ -606,8 +727,12 @@ int InstrEmit_stdux(PPCHIRBuilder& f, const InstrData& i) {
// MEM(EA, 8) <- (RS)
// RA <- EA
Value* ea = CalculateEA(f, i.X.RA, i.X.RB);
- f.Store(ea, f.ByteSwap(f.LoadGPR(i.X.RT)));
+ Value* val64 = f.LoadGPR(i.X.RT);
+ f.Store(ea, f.ByteSwap(val64));
StoreEA(f, i.X.RA, ea);
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch64(f, i.address, val64, ea);
+ }
return 0;
}
@@ -619,7 +744,11 @@ int InstrEmit_stdx(PPCHIRBuilder& f, const InstrData& i) {
// EA <- b + (RB)
// MEM(EA, 8) <- (RS)
Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
- f.Store(ea, f.ByteSwap(f.LoadGPR(i.X.RT)));
+ Value* val64 = f.LoadGPR(i.X.RT);
+ f.Store(ea, f.ByteSwap(val64));
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch64(f, i.address, val64, ea);
+ }
return 0;
}
@@ -684,7 +813,11 @@ int InstrEmit_stwbrx(PPCHIRBuilder& f, const InstrData& i) {
// EA <- b + (RB)
// MEM(EA, 4) <- bswap((RS)[32:63])
Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
- f.Store(ea, f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE));
+ Value* val32 = f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE);
+ f.Store(ea, val32);
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch(f, i.address, val32, ea);
+ }
return 0;
}
@@ -696,7 +829,11 @@ int InstrEmit_stdbrx(PPCHIRBuilder& f, const InstrData& i) {
// EA <- b + (RB)
// MEM(EA, 8) <- bswap(RS)
Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
- f.Store(ea, f.LoadGPR(i.X.RT));
+ Value* val64 = f.LoadGPR(i.X.RT);
+ f.Store(ea, val64);
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch64(f, i.address, val64, ea);
+ }
return 0;
}
@@ -843,7 +980,8 @@ int InstrEmit_stdcx(PPCHIRBuilder& f, const InstrData& i) {
// This will always succeed if under the global lock, however.
Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
- Value* rt = f.ByteSwap(f.LoadGPR(i.X.RT));
+ Value* val64 = f.LoadGPR(i.X.RT);
+ Value* rt = f.ByteSwap(val64);
if (cvars::no_reserved_ops) {
f.Store(ea, rt);
@@ -862,6 +1000,9 @@ int InstrEmit_stdcx(PPCHIRBuilder& f, const InstrData& i) {
if (!cvars::no_reserved_ops) {
f.MemoryBarrier();
}
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch64(f, i.address, val64, ea);
+ }
return 0;
}
@@ -885,7 +1026,8 @@ int InstrEmit_stwcx(PPCHIRBuilder& f, const InstrData& i) {
Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
- Value* rt = f.ByteSwap(f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE));
+ Value* val32 = f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE);
+ Value* rt = f.ByteSwap(val32);
if (cvars::no_reserved_ops) {
f.Store(ea, rt);
@@ -904,7 +1046,9 @@ int InstrEmit_stwcx(PPCHIRBuilder& f, const InstrData& i) {
if (!cvars::no_reserved_ops) {
f.MemoryBarrier();
}
-
+ if (!::xe::cpu::audit67::vals().empty()) {
+ EmitAudit67ValueWatch(f, i.address, val32, ea);
+ }
return 0;
}
// Floating-point load (A-19)
diff --git a/src/xenia/cpu/ppc/ppc_hir_builder.cc b/src/xenia/cpu/ppc/ppc_hir_builder.cc
index 42d996cba..e2f7a45db 100644
--- a/src/xenia/cpu/ppc/ppc_hir_builder.cc
+++ b/src/xenia/cpu/ppc/ppc_hir_builder.cc
@@ -34,6 +34,97 @@ DEFINE_bool(
"unimplemented PowerPC instruction is encountered.",
"CPU");
+// AUDIT-061 — multi-PC branch probe. Parses cvars::audit_61_branch_probe_pcs
+// once and exposes a (pc -> trap_id) lookup table. trap_id range [200, 65535].
+// PCs outside the table are not probed. Native side reads g_audit61_pcs[idx].
+#include <vector>
+#include <string>
+namespace xe {
+namespace cpu {
+namespace audit61 {
+constexpr uint16_t kTrapBase = 200;
+constexpr size_t kMaxPcs = 32;
+static std::vector<uint32_t> g_pcs;
+static bool g_parsed = false;
+
+const std::vector<uint32_t>& pcs() {
+ if (!g_parsed) {
+ g_parsed = true;
+ const std::string& csv = cvars::audit_61_branch_probe_pcs;
+ size_t pos = 0;
+ while (pos < csv.size() && g_pcs.size() < kMaxPcs) {
+ size_t end = csv.find(',', pos);
+ std::string tok = csv.substr(pos, end - pos);
+ // strip whitespace
+ while (!tok.empty() && (tok.front() == ' ' || tok.front() == '\t'))
+ tok.erase(tok.begin());
+ while (!tok.empty() && (tok.back() == ' ' || tok.back() == '\t'))
+ tok.pop_back();
+ if (!tok.empty()) {
+ try {
+ uint32_t v = static_cast<uint32_t>(std::stoul(tok, nullptr, 0));
+ g_pcs.push_back(v);
+ } catch (...) {
+ }
+ }
+ if (end == std::string::npos) break;
+ pos = end + 1;
+ }
+ }
+ return g_pcs;
+}
+
+// Returns trap id for pc, or 0 if pc not in probe set.
+uint16_t trap_id_for(uint32_t pc) {
+ const auto& v = pcs();
+ for (size_t i = 0; i < v.size(); ++i) {
+ if (v[i] == pc) return static_cast<uint16_t>(kTrapBase + i);
+ }
+ return 0;
+}
+} // namespace audit61
+
+// AUDIT-067 — value-watch. Parses cvars::audit_67_value_watch once, exposes
+// values via vals(). Trap codes for matches start at kTrapBase = 250.
+namespace audit67 {
+constexpr uint16_t kTrapBase = 250;
+constexpr size_t kMaxVals = 4;
+static std::vector<uint32_t> g_vals;
+static bool g_parsed = false;
+
+const std::vector<uint32_t>& vals() {
+ if (!g_parsed) {
+ g_parsed = true;
+ const std::string& csv = cvars::audit_67_value_watch;
+ size_t pos = 0;
+ while (pos < csv.size() && g_vals.size() < kMaxVals) {
+ size_t end = csv.find(',', pos);
+ std::string tok = csv.substr(pos, end - pos);
+ while (!tok.empty() && (tok.front() == ' ' || tok.front() == '\t'))
+ tok.erase(tok.begin());
+ while (!tok.empty() && (tok.back() == ' ' || tok.back() == '\t'))
+ tok.pop_back();
+ if (!tok.empty()) {
+ try {
+ uint32_t v = static_cast<uint32_t>(std::stoul(tok, nullptr, 0));
+ g_vals.push_back(v);
+ } catch (...) {
+ }
+ }
+ if (end == std::string::npos) break;
+ pos = end + 1;
+ }
+ XELOGI("AUDIT-067-INIT csv=\"{}\" parsed_count={}", csv, g_vals.size());
+ for (size_t i = 0; i < g_vals.size(); ++i) {
+ XELOGI("AUDIT-067-INIT vals[{}] = 0x{:08X}", i, g_vals[i]);
+ }
+ }
+ return g_vals;
+}
+} // namespace audit67
+} // namespace cpu
+} // namespace xe
+
namespace xe {
namespace cpu {
namespace ppc {
@@ -174,6 +265,20 @@ bool PPCHIRBuilder::Emit(GuestFunction* function, uint32_t flags) {
MaybeBreakOnInstruction(address);
+ // AUDIT-061: emit a trap before this instruction if it's on the probe
+ // list. The trap fires BEFORE the cmp/branch HIR emit so the native
+ // handler observes cr0/cr6 set by the *previous* instruction (the cmp
+ // that controls this conditional branch). ContextBarrier flushes
+ // HIR temporaries to PPCContext so the handler reads consistent state.
+ if (!::xe::cpu::audit61::pcs().empty()) {
+ uint16_t tid = ::xe::cpu::audit61::trap_id_for(address);
+ if (tid != 0) {
+ Comment("--audit_61_branch_probe target");
+ ContextBarrier();
+ Trap(tid);
+ }
+ }
+
InstrData i;
i.address = address;
i.code = code;