Removing win32 code from X64CodeCache.

Fixes #349.
This commit is contained in:
Ben Vanik
2015-07-19 10:33:00 -07:00
parent edfa3f3fc0
commit 352bae30cb
5 changed files with 326 additions and 228 deletions

View File

@@ -18,19 +18,11 @@
#include "xenia/base/math.h"
#include "xenia/base/memory.h"
// When enabled, this will use Windows 8 APIs to get unwind info.
// TODO(benvanik): figure out why the callback variant doesn't work.
#define USE_GROWABLE_FUNCTION_TABLE
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
// Size of unwind info per function.
// TODO(benvanik): move this to emitter.
const static uint32_t kUnwindInfoSize = 4 + (2 * 1 + 2 + 2);
X64CodeCache::X64CodeCache() = default;
X64CodeCache::~X64CodeCache() {
@@ -39,17 +31,6 @@ X64CodeCache::~X64CodeCache() {
xe::memory::DeallocationType::kRelease);
}
#ifdef USE_GROWABLE_FUNCTION_TABLE
if (unwind_table_handle_) {
RtlDeleteGrowableFunctionTable(unwind_table_handle_);
}
#else
if (generated_code_base_) {
RtlDeleteFunctionTable(reinterpret_cast<PRUNTIME_FUNCTION>(
reinterpret_cast<DWORD64>(generated_code_base_) | 0x3));
}
#endif // USE_GROWABLE_FUNCTION_TABLE
// Unmap all views and close mapping.
if (mapping_) {
xe::memory::UnmapFileView(mapping_, generated_code_base_,
@@ -97,38 +78,6 @@ bool X64CodeCache::Initialize() {
return false;
}
// Compute total number of unwind entries we should allocate.
// We don't support reallocing right now, so this should be high.
unwind_table_.resize(30000);
#ifdef USE_GROWABLE_FUNCTION_TABLE
// Create table and register with the system. It's empty now, but we'll grow
// it as functions are added.
if (RtlAddGrowableFunctionTable(
&unwind_table_handle_, unwind_table_.data(), unwind_table_count_,
DWORD(unwind_table_.size()),
reinterpret_cast<ULONG_PTR>(generated_code_base_),
reinterpret_cast<ULONG_PTR>(generated_code_base_ +
kGeneratedCodeSize))) {
XELOGE("Unable to create unwind function table");
return false;
}
#else
// Install a callback that the debugger will use to lookup unwind info on
// demand.
if (!RtlInstallFunctionTableCallback(
reinterpret_cast<DWORD64>(generated_code_base_) | 0x3,
reinterpret_cast<DWORD64>(generated_code_base_),
kGeneratedCodeSize, [](uintptr_t control_pc, void* context) {
auto code_cache = reinterpret_cast<X64CodeCache*>(context);
return reinterpret_cast<PRUNTIME_FUNCTION>(
code_cache->LookupUnwindEntry(control_pc));
}, this, nullptr)) {
XELOGE("Unable to install function table callback");
return false;
}
#endif // USE_GROWABLE_FUNCTION_TABLE
return true;
}
@@ -165,8 +114,7 @@ void* X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
size_t low_mark;
size_t high_mark;
uint8_t* code_address = nullptr;
uint8_t* unwind_entry_address = nullptr;
size_t unwind_table_slot = 0;
UnwindReservation unwind_reservation;
{
std::lock_guard<xe::mutex> allocation_lock(allocation_mutex_);
@@ -180,9 +128,9 @@ void* X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
// Reserve unwind info.
// We go on the high size of the unwind info as we don't know how big we
// need it, and a few extra bytes of padding isn't the worst thing.
unwind_entry_address = generated_code_base_ + generated_code_offset_;
generated_code_offset_ += xe::round_up(kUnwindInfoSize, 16);
unwind_table_slot = ++unwind_table_count_;
unwind_reservation =
RequestUnwindReservation(generated_code_base_ + generated_code_offset_);
generated_code_offset_ += unwind_reservation.data_size;
high_mark = generated_code_offset_;
}
@@ -203,18 +151,9 @@ void* X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
// Copy code.
std::memcpy(code_address, machine_code, code_size);
// Add unwind info.
InitializeUnwindEntry(unwind_entry_address, unwind_table_slot, code_address,
code_size, stack_size);
#ifdef USE_GROWABLE_FUNCTION_TABLE
// Notify that the unwind table has grown.
// We do this outside of the lock, but with the latest total count.
RtlGrowFunctionTable(unwind_table_handle_, unwind_table_count_);
#endif // USE_GROWABLE_FUNCTION_TABLE
// This isn't needed on x64 (probably), but is convention.
FlushInstructionCache(GetCurrentProcess(), code_address, code_size);
// Notify subclasses of placed code.
PlaceCode(guest_address, machine_code, code_size, stack_size, code_address,
unwind_reservation);
// Now that everything is ready, fix up the indirection table.
// Note that we do support code that doesn't have an indirection fixup, so
@@ -228,146 +167,6 @@ void* X64CodeCache::PlaceCode(uint32_t guest_address, void* machine_code,
return code_address;
}
// http://msdn.microsoft.com/en-us/library/ssa62fwe.aspx
typedef enum _UNWIND_OP_CODES {
UWOP_PUSH_NONVOL = 0, /* info == register number */
UWOP_ALLOC_LARGE, /* no info, alloc size in next 2 slots */
UWOP_ALLOC_SMALL, /* info == size of allocation / 8 - 1 */
UWOP_SET_FPREG, /* no info, FP = RSP + UNWIND_INFO.FPRegOffset*16 */
UWOP_SAVE_NONVOL, /* info == register number, offset in next slot */
UWOP_SAVE_NONVOL_FAR, /* info == register number, offset in next 2 slots */
UWOP_SAVE_XMM128, /* info == XMM reg number, offset in next slot */
UWOP_SAVE_XMM128_FAR, /* info == XMM reg number, offset in next 2 slots */
UWOP_PUSH_MACHFRAME /* info == 0: no error-code, 1: error-code */
} UNWIND_CODE_OPS;
class UNWIND_REGISTER {
public:
enum _ {
RAX = 0,
RCX = 1,
RDX = 2,
RBX = 3,
RSP = 4,
RBP = 5,
RSI = 6,
RDI = 7,
R8 = 8,
R9 = 9,
R10 = 10,
R11 = 11,
R12 = 12,
R13 = 13,
R14 = 14,
R15 = 15,
};
};
typedef union _UNWIND_CODE {
struct {
uint8_t CodeOffset;
uint8_t UnwindOp : 4;
uint8_t OpInfo : 4;
};
USHORT FrameOffset;
} UNWIND_CODE, *PUNWIND_CODE;
typedef struct _UNWIND_INFO {
uint8_t Version : 3;
uint8_t Flags : 5;
uint8_t SizeOfProlog;
uint8_t CountOfCodes;
uint8_t FrameRegister : 4;
uint8_t FrameOffset : 4;
UNWIND_CODE UnwindCode[1];
/* UNWIND_CODE MoreUnwindCode[((CountOfCodes + 1) & ~1) - 1];
* union {
* OPTIONAL ULONG ExceptionHandler;
* OPTIONAL ULONG FunctionEntry;
* };
* OPTIONAL ULONG ExceptionData[]; */
} UNWIND_INFO, *PUNWIND_INFO;
void X64CodeCache::InitializeUnwindEntry(uint8_t* unwind_entry_address,
size_t unwind_table_slot,
uint8_t* code_address,
size_t code_size, size_t stack_size) {
auto unwind_info = reinterpret_cast<UNWIND_INFO*>(unwind_entry_address);
if (!stack_size) {
// http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
unwind_info->Version = 1;
unwind_info->Flags = 0;
unwind_info->SizeOfProlog = 0;
unwind_info->CountOfCodes = 0;
unwind_info->FrameRegister = 0;
unwind_info->FrameOffset = 0;
} else if (stack_size <= 128) {
uint8_t prolog_size = 4;
// http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
unwind_info->Version = 1;
unwind_info->Flags = 0;
unwind_info->SizeOfProlog = prolog_size;
unwind_info->CountOfCodes = 1;
unwind_info->FrameRegister = 0;
unwind_info->FrameOffset = 0;
// http://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
size_t co = 0;
auto& unwind_code = unwind_info->UnwindCode[co++];
unwind_code.CodeOffset =
14; // end of instruction + 1 == offset of next instruction
unwind_code.UnwindOp = UWOP_ALLOC_SMALL;
unwind_code.OpInfo = stack_size / 8 - 1;
} else {
// TODO(benvanik): take as parameters?
uint8_t prolog_size = 7;
// http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
unwind_info->Version = 1;
unwind_info->Flags = 0;
unwind_info->SizeOfProlog = prolog_size;
unwind_info->CountOfCodes = 2;
unwind_info->FrameRegister = 0;
unwind_info->FrameOffset = 0;
// http://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
size_t co = 0;
auto& unwind_code = unwind_info->UnwindCode[co++];
unwind_code.CodeOffset =
7; // end of instruction + 1 == offset of next instruction
unwind_code.UnwindOp = UWOP_ALLOC_LARGE;
unwind_code.OpInfo = 0;
unwind_code = unwind_info->UnwindCode[co++];
unwind_code.FrameOffset = (USHORT)(stack_size) / 8;
}
// Add entry.
auto& fn_entry = unwind_table_[unwind_table_slot];
fn_entry.BeginAddress = (DWORD)(code_address - generated_code_base_);
fn_entry.EndAddress = (DWORD)(fn_entry.BeginAddress + code_size);
fn_entry.UnwindData = (DWORD)(unwind_entry_address - generated_code_base_);
}
void* X64CodeCache::LookupUnwindEntry(uintptr_t host_address) {
void* fn_entry = std::bsearch(
&host_address, unwind_table_.data(), unwind_table_count_ + 1,
sizeof(RUNTIME_FUNCTION),
[](const void* key_ptr, const void* element_ptr) {
auto key =
*reinterpret_cast<const uintptr_t*>(key_ptr) - kGeneratedCodeBase;
auto element = reinterpret_cast<const RUNTIME_FUNCTION*>(element_ptr);
if (key < element->BeginAddress) {
return -1;
} else if (key > element->EndAddress) {
return 1;
} else {
return 0;
}
});
return reinterpret_cast<RUNTIME_FUNCTION*>(fn_entry);
}
uint32_t X64CodeCache::PlaceData(const void* data, size_t length) {
// Hold a lock while we bump the pointers up.
size_t high_mark;