Mostly complete tracing. Probably a lot of bugs.

This commit is contained in:
Ben Vanik
2014-08-14 20:28:44 -07:00
parent cebf595958
commit c275562594
54 changed files with 1052 additions and 161 deletions

View File

@@ -43,6 +43,7 @@ class Assembler {
virtual int Assemble(runtime::FunctionInfo* symbol_info,
hir::HIRBuilder* builder, uint32_t debug_info_flags,
std::unique_ptr<runtime::DebugInfo> debug_info,
uint32_t trace_flags,
runtime::Function** out_function) = 0;
protected:

View File

@@ -50,7 +50,7 @@ void IVMAssembler::Reset() {
int IVMAssembler::Assemble(FunctionInfo* symbol_info, HIRBuilder* builder,
uint32_t debug_info_flags,
std::unique_ptr<DebugInfo> debug_info,
Function** out_function) {
uint32_t trace_flags, Function** out_function) {
SCOPE_profile_cpu_f("alloy");
// Reset when we leave.

View File

@@ -31,7 +31,7 @@ class IVMAssembler : public Assembler {
int Assemble(runtime::FunctionInfo* symbol_info, hir::HIRBuilder* builder,
uint32_t debug_info_flags,
std::unique_ptr<runtime::DebugInfo> debug_info,
runtime::Function** out_function) override;
uint32_t trace_flags, runtime::Function** out_function) override;
private:
Arena intcode_arena_;

View File

@@ -11,11 +11,12 @@
#include <algorithm>
#include <poly/poly.h>
#include <alloy/hir/label.h>
#include <alloy/runtime/runtime.h>
#include <alloy/runtime/symbol_info.h>
#include <alloy/runtime/thread_state.h>
#include <poly/poly.h>
#include <xdb/protocol.h>
// TODO(benvanik): make a compile time flag?
//#define DYNAMIC_REGISTER_ACCESS_CHECK(address) false
@@ -246,6 +247,22 @@ int Translate_SOURCE_OFFSET(TranslationContext& ctx, Instr* i) {
return 0;
}
// TODO(benvanik): dispatch of register forms.
uint32_t IntCode_TRACE_SOURCE(IntCodeState& ics, const IntCode* i) {
// TODO(benvanik): append to active trace writer.
uint64_t trace_base = ics.thread_state->memory()->trace_base();
if (trace_base) {
auto ev = xdb::protocol::InstrEvent::Append(trace_base);
ev->type = xdb::protocol::EventType::INSTR;
ev->thread_id = ics.thread_state->thread_id();
ev->address = ics.rf[i->src1_reg].i32;
}
return IA_NEXT;
}
int Translate_TRACE_SOURCE(TranslationContext& ctx, Instr* i) {
return DispatchToC(ctx, i, IntCode_TRACE_SOURCE);
}
uint32_t IntCode_DEBUG_BREAK(IntCodeState& ics, const IntCode* i) {
DFLUSH();
__debugbreak();
@@ -4174,33 +4191,33 @@ int Translate_ATOMIC_EXCHANGE(TranslationContext& ctx, Instr* i) {
typedef int (*TranslateFn)(TranslationContext& ctx, Instr* i);
static const TranslateFn dispatch_table[] = {
Translate_COMMENT, Translate_NOP,
Translate_SOURCE_OFFSET, Translate_DEBUG_BREAK,
Translate_DEBUG_BREAK_TRUE, Translate_TRAP,
Translate_TRAP_TRUE, Translate_CALL,
Translate_CALL_TRUE, Translate_CALL_INDIRECT,
Translate_CALL_INDIRECT_TRUE, Translate_CALL_EXTERN,
Translate_RETURN, Translate_RETURN_TRUE,
Translate_SET_RETURN_ADDRESS, Translate_BRANCH,
Translate_BRANCH_TRUE, Translate_BRANCH_FALSE,
Translate_ASSIGN, Translate_CAST,
Translate_ZERO_EXTEND, Translate_SIGN_EXTEND,
Translate_TRUNCATE, Translate_CONVERT,
Translate_ROUND, Translate_VECTOR_CONVERT_I2F,
Translate_VECTOR_CONVERT_F2I, Translate_LOAD_VECTOR_SHL,
Translate_LOAD_VECTOR_SHR, Translate_LOAD_CLOCK,
Translate_LOAD_LOCAL, Translate_STORE_LOCAL,
Translate_LOAD_CONTEXT, Translate_STORE_CONTEXT,
Translate_LOAD, Translate_STORE,
Translate_PREFETCH, Translate_MAX,
Translate_VECTOR_MAX, Translate_MIN,
Translate_VECTOR_MIN, Translate_SELECT,
Translate_IS_TRUE, Translate_IS_FALSE,
Translate_COMPARE_EQ, Translate_COMPARE_NE,
Translate_COMPARE_SLT, Translate_COMPARE_SLE,
Translate_COMPARE_SGT, Translate_COMPARE_SGE,
Translate_COMPARE_ULT, Translate_COMPARE_ULE,
Translate_COMPARE_UGT, Translate_COMPARE_UGE,
Translate_DID_CARRY,
Translate_SOURCE_OFFSET, Translate_TRACE_SOURCE,
Translate_DEBUG_BREAK, Translate_DEBUG_BREAK_TRUE,
Translate_TRAP, Translate_TRAP_TRUE,
Translate_CALL, Translate_CALL_TRUE,
Translate_CALL_INDIRECT, Translate_CALL_INDIRECT_TRUE,
Translate_CALL_EXTERN, Translate_RETURN,
Translate_RETURN_TRUE, Translate_SET_RETURN_ADDRESS,
Translate_BRANCH, Translate_BRANCH_TRUE,
Translate_BRANCH_FALSE, Translate_ASSIGN,
Translate_CAST, Translate_ZERO_EXTEND,
Translate_SIGN_EXTEND, Translate_TRUNCATE,
Translate_CONVERT, Translate_ROUND,
Translate_VECTOR_CONVERT_I2F, Translate_VECTOR_CONVERT_F2I,
Translate_LOAD_VECTOR_SHL, Translate_LOAD_VECTOR_SHR,
Translate_LOAD_CLOCK, Translate_LOAD_LOCAL,
Translate_STORE_LOCAL, Translate_LOAD_CONTEXT,
Translate_STORE_CONTEXT, Translate_LOAD,
Translate_STORE, Translate_PREFETCH,
Translate_MAX, Translate_VECTOR_MAX,
Translate_MIN, Translate_VECTOR_MIN,
Translate_SELECT, Translate_IS_TRUE,
Translate_IS_FALSE, Translate_COMPARE_EQ,
Translate_COMPARE_NE, Translate_COMPARE_SLT,
Translate_COMPARE_SLE, Translate_COMPARE_SGT,
Translate_COMPARE_SGE, Translate_COMPARE_ULT,
Translate_COMPARE_ULE, Translate_COMPARE_UGT,
Translate_COMPARE_UGE, Translate_DID_CARRY,
TranslateInvalid, // Translate_DID_OVERFLOW,
Translate_DID_SATURATE, Translate_VECTOR_COMPARE_EQ,
Translate_VECTOR_COMPARE_SGT, Translate_VECTOR_COMPARE_SGE,

View File

@@ -8,6 +8,7 @@
'ivm_backend.cc',
'ivm_backend.h',
'ivm_function.cc',
'ivm_function.h',
'ivm_stack.cc',
'ivm_stack.h',
],

View File

@@ -58,7 +58,7 @@ void X64Assembler::Reset() {
int X64Assembler::Assemble(FunctionInfo* symbol_info, HIRBuilder* builder,
uint32_t debug_info_flags,
std::unique_ptr<DebugInfo> debug_info,
Function** out_function) {
uint32_t trace_flags, Function** out_function) {
SCOPE_profile_cpu_f("alloy");
// Reset when we leave.
@@ -68,7 +68,7 @@ int X64Assembler::Assemble(FunctionInfo* symbol_info, HIRBuilder* builder,
void* machine_code = 0;
size_t code_size = 0;
int result = emitter_->Emit(builder, debug_info_flags, debug_info.get(),
machine_code, code_size);
trace_flags, machine_code, code_size);
if (result) {
return result;
}

View File

@@ -37,7 +37,7 @@ class X64Assembler : public Assembler {
int Assemble(runtime::FunctionInfo* symbol_info, hir::HIRBuilder* builder,
uint32_t debug_info_flags,
std::unique_ptr<runtime::DebugInfo> debug_info,
runtime::Function** out_function) override;
uint32_t trace_flags, runtime::Function** out_function) override;
private:
void DumpMachineCode(runtime::DebugInfo* debug_info, void* machine_code,

View File

@@ -19,6 +19,7 @@
#include <alloy/runtime/runtime.h>
#include <alloy/runtime/symbol_info.h>
#include <alloy/runtime/thread_state.h>
#include <xdb/protocol.h>
namespace alloy {
namespace backend {
@@ -66,8 +67,8 @@ X64Emitter::~X64Emitter() {}
int X64Emitter::Initialize() { return 0; }
int X64Emitter::Emit(HIRBuilder* builder, uint32_t debug_info_flags,
runtime::DebugInfo* debug_info, void*& out_code_address,
size_t& out_code_size) {
runtime::DebugInfo* debug_info, uint32_t trace_flags,
void*& out_code_address, size_t& out_code_size) {
SCOPE_profile_cpu_f("alloy");
// Reset.
@@ -75,6 +76,7 @@ int X64Emitter::Emit(HIRBuilder* builder, uint32_t debug_info_flags,
source_map_count_ = 0;
source_map_arena_.Reset();
}
trace_flags_ = trace_flags;
// Fill the generator with code.
size_t stack_size = 0;
@@ -151,6 +153,19 @@ int X64Emitter::Emit(HIRBuilder* builder, size_t& out_stack_size) {
mov(rdx, qword[rcx + 8]); // membase
}
uint64_t trace_base = runtime_->memory()->trace_base();
if (trace_base && trace_flags_ & TRACE_USER_CALLS) {
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(sizeof(xdb::protocol::UserCallEvent)));
lock();
xadd(qword[rax], r8);
mov(rax, static_cast<uint64_t>(xdb::protocol::EventType::USER_CALL) |
(static_cast<uint64_t>(0) << 8) | (0ull << 32));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
}
// Body.
auto block = builder->first_block();
while (block) {
@@ -165,6 +180,16 @@ int X64Emitter::Emit(HIRBuilder* builder, size_t& out_stack_size) {
const Instr* instr = block->instr_head;
while (instr) {
const Instr* new_tail = instr;
// Special handling of TRACE_SOURCE.
if (instr->opcode == &OPCODE_TRACE_SOURCE_info) {
if (trace_flags_ & TRACE_SOURCE) {
EmitTraceSource(instr);
}
instr = instr->next;
continue;
}
if (!SelectSequence(*this, instr, &new_tail)) {
// No sequence found!
assert_always();
@@ -179,6 +204,7 @@ int X64Emitter::Emit(HIRBuilder* builder, size_t& out_stack_size) {
// Function epilog.
L("epilog");
EmitTraceUserCallReturn();
if (emit_prolog) {
mov(rcx, qword[rsp + StackLayout::GUEST_RCX_HOME]);
add(rsp, (uint32_t)stack_size);
@@ -204,6 +230,129 @@ void X64Emitter::MarkSourceOffset(const Instr* i) {
source_map_count_++;
}
void X64Emitter::EmitTraceSource(const Instr* instr) {
uint64_t trace_base = runtime_->memory()->trace_base();
if (!trace_base || !(trace_flags_ & TRACE_SOURCE)) {
return;
}
// TODO(benvanik): make this a function call to some append fn.
uint8_t dest_reg_0 = instr->flags & 0xFF;
uint8_t dest_reg_1 = instr->flags >> 8;
xdb::protocol::EventType event_type;
size_t event_size;
if (dest_reg_0 == 100) {
event_type = xdb::protocol::EventType::INSTR;
event_size = sizeof(xdb::protocol::InstrEvent);
} else if (dest_reg_1 == 100) {
if (dest_reg_0 & (1 << 7)) {
event_type = xdb::protocol::EventType::INSTR_R16;
event_size = sizeof(xdb::protocol::InstrEventR16);
} else {
event_type = xdb::protocol::EventType::INSTR_R8;
event_size = sizeof(xdb::protocol::InstrEventR8);
}
} else {
if (dest_reg_0 & (1 << 7) && dest_reg_1 & (1 << 7)) {
event_type = xdb::protocol::EventType::INSTR_R16_R16;
event_size = sizeof(xdb::protocol::InstrEventR16R16);
} else if (dest_reg_0 & (1 << 7) && !(dest_reg_1 & (1 << 7))) {
event_type = xdb::protocol::EventType::INSTR_R16_R8;
event_size = sizeof(xdb::protocol::InstrEventR16R8);
} else if (!(dest_reg_0 & (1 << 7)) && dest_reg_1 & (1 << 7)) {
event_type = xdb::protocol::EventType::INSTR_R8_R16;
event_size = sizeof(xdb::protocol::InstrEventR8R16);
} else if (!(dest_reg_0 & (1 << 7)) && !(dest_reg_1 & (1 << 7))) {
event_type = xdb::protocol::EventType::INSTR_R8_R8;
event_size = sizeof(xdb::protocol::InstrEventR8R8);
}
}
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(event_size));
lock();
xadd(qword[rax], r8);
// r8 is now the pointer where we can write our event.
// Write the header, which is the same for everything (pretty much).
// Some event types ignore the dest reg, and that's fine.
uint64_t qword_0 = static_cast<uint64_t>(event_type) |
(static_cast<uint64_t>(dest_reg_0) << 8) |
(instr->src1.offset << 32);
mov(rax, qword_0);
mov(qword[r8], rax);
// Write thread ID.
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
switch (event_type) {
case xdb::protocol::EventType::INSTR:
break;
case xdb::protocol::EventType::INSTR_R8:
case xdb::protocol::EventType::INSTR_R16:
if (dest_reg_0 & (1 << 7)) {
EmitTraceSourceAppendValue(instr->src2.value, 8);
} else {
EmitTraceSourceAppendValue(instr->src2.value, 8);
}
break;
case xdb::protocol::EventType::INSTR_R8_R8:
case xdb::protocol::EventType::INSTR_R8_R16:
case xdb::protocol::EventType::INSTR_R16_R8:
case xdb::protocol::EventType::INSTR_R16_R16:
mov(word[r8 + 8], dest_reg_0 | static_cast<uint16_t>(dest_reg_1 << 8));
size_t offset = 8;
if (dest_reg_0 & (1 << 7)) {
EmitTraceSourceAppendValue(instr->src2.value, offset);
offset += 16;
} else {
EmitTraceSourceAppendValue(instr->src2.value, offset);
offset += 8;
}
if (dest_reg_1 & (1 << 7)) {
EmitTraceSourceAppendValue(instr->src3.value, offset);
offset += 16;
} else {
EmitTraceSourceAppendValue(instr->src3.value, offset);
offset += 8;
}
break;
}
}
void X64Emitter::EmitTraceSourceAppendValue(const Value* value,
size_t r8_offset) {
//
}
void X64Emitter::EmitGetCurrentThreadId() {
// rcx must point to context. We could fetch from the stack if needed.
mov(ax,
word[rcx + runtime_->frontend()->context_info()->thread_id_offset()]);
}
void X64Emitter::EmitTraceUserCallReturn() {
auto trace_base = runtime_->memory()->trace_base();
if (!trace_base || !(trace_flags_ & TRACE_USER_CALLS)) {
return;
}
mov(rdx, rax);
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(sizeof(xdb::protocol::UserCallReturnEvent)));
lock();
xadd(qword[rax], r8);
mov(rax, static_cast<uint64_t>(xdb::protocol::EventType::USER_CALL_RETURN) |
(static_cast<uint64_t>(0) << 8) | (0ull << 32));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
mov(rax, rdx);
ReloadEDX();
}
void X64Emitter::DebugBreak() {
// TODO(benvanik): notify debugger.
db(0xCC);
@@ -258,7 +407,7 @@ uint64_t ResolveFunctionSymbol(void* raw_context, uint64_t symbol_info_ptr) {
#else
uint64_t return_address =
reinterpret_cast<uint64_t>(__builtin_return_address(0));
#endif // XE_WIN32_LIKE
#endif // XE_LIKE_WIN32
#pragma pack(push, 1)
struct Asm {
uint16_t mov_rax;
@@ -301,6 +450,9 @@ void X64Emitter::Call(const hir::Instr* instr,
// Actually jump/call to rax.
if (instr->flags & CALL_TAIL) {
// Since we skip the prolog we need to mark the return here.
EmitTraceUserCallReturn();
// Pass the callers return address over.
mov(rdx, qword[rsp + StackLayout::GUEST_RET_ADDR]);
@@ -338,7 +490,7 @@ uint64_t ResolveFunctionAddress(void* raw_context, uint64_t target_address) {
#else
uint64_t return_address =
reinterpret_cast<uint64_t>(__builtin_return_address(0));
#endif // XE_WIN32_LIKE
#endif // XE_LIKE_WIN32
#pragma pack(push, 1)
struct Asm {
uint16_t cmp_rdx;
@@ -426,6 +578,9 @@ void X64Emitter::CallIndirect(const hir::Instr* instr, const Reg64& reg) {
// Actually jump/call to rax.
L(skip_resolve);
if (instr->flags & CALL_TAIL) {
// Since we skip the prolog we need to mark the return here.
EmitTraceUserCallReturn();
// Pass the callers return address over.
mov(rdx, qword[rsp + StackLayout::GUEST_RET_ADDR]);
@@ -449,6 +604,24 @@ uint64_t UndefinedCallExtern(void* raw_context, uint64_t symbol_info_ptr) {
void X64Emitter::CallExtern(const hir::Instr* instr,
const FunctionInfo* symbol_info) {
assert_true(symbol_info->behavior() == FunctionInfo::BEHAVIOR_EXTERN);
uint64_t trace_base = runtime_->memory()->trace_base();
if (trace_base & trace_flags_ & TRACE_EXTERN_CALLS) {
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(sizeof(xdb::protocol::KernelCallEvent)));
lock();
xadd(qword[rax], r8);
// TODO(benvanik): get module/ordinal.
uint32_t module_id = 0;
uint32_t ordinal = 0;
mov(rax, static_cast<uint64_t>(xdb::protocol::EventType::KERNEL_CALL) |
(static_cast<uint64_t>(0) << 8) | (module_id << 16) |
(ordinal));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
}
if (!symbol_info->extern_handler()) {
CallNative(UndefinedCallExtern, reinterpret_cast<uint64_t>(symbol_info));
} else {
@@ -466,6 +639,21 @@ void X64Emitter::CallExtern(const hir::Instr* instr,
ReloadEDX();
// rax = host return
}
if (trace_base && trace_flags_ & TRACE_EXTERN_CALLS) {
mov(rax, trace_base);
mov(r8d,
static_cast<uint32_t>(sizeof(xdb::protocol::KernelCallReturnEvent)));
lock();
xadd(qword[rax], r8);
uint32_t module_id = 0;
uint32_t ordinal = 0;
mov(rax,
static_cast<uint64_t>(xdb::protocol::EventType::KERNEL_CALL_RETURN) |
(static_cast<uint64_t>(0) << 8) | (0));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
}
}
void X64Emitter::CallNative(void* fn) {

View File

@@ -84,8 +84,8 @@ class X64Emitter : public Xbyak::CodeGenerator {
int Initialize();
int Emit(hir::HIRBuilder* builder, uint32_t debug_info_flags,
runtime::DebugInfo* debug_info, void*& out_code_address,
size_t& out_code_size);
runtime::DebugInfo* debug_info, uint32_t trace_flags,
void*& out_code_address, size_t& out_code_size);
public:
// Reserved: rsp
@@ -163,6 +163,10 @@ class X64Emitter : public Xbyak::CodeGenerator {
protected:
void* Emplace(size_t stack_size);
int Emit(hir::HIRBuilder* builder, size_t& out_stack_size);
void EmitTraceSource(const hir::Instr* instr);
void EmitTraceSourceAppendValue(const hir::Value* value, size_t r8_offset);
void EmitGetCurrentThreadId();
void EmitTraceUserCallReturn();
protected:
runtime::Runtime* runtime_;
@@ -177,6 +181,8 @@ class X64Emitter : public Xbyak::CodeGenerator {
size_t stack_size_;
uint32_t trace_flags_;
static const uint32_t gpr_reg_map_[GPR_COUNT];
static const uint32_t xmm_reg_map_[XMM_COUNT];
};