1
0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-21 18:22:33 +01:00

Reservations improved

This commit is contained in:
Nekotekina 2015-02-17 03:08:23 +03:00
parent e6c628caba
commit 20dce26b1d
4 changed files with 313 additions and 72 deletions

View File

@ -51,7 +51,7 @@ void SetCurrentThreadDebugName(const char* threadName)
enum x64_reg_t : u32
{
X64R_RAX,
X64R_RAX = 0,
X64R_RCX,
X64R_RDX,
X64R_RBX,
@ -68,7 +68,7 @@ enum x64_reg_t : u32
X64R_R14,
X64R_R15,
X64R_XMM0,
X64R_XMM0 = 0,
X64R_XMM1,
X64R_XMM2,
X64R_XMM3,
@ -140,7 +140,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
{
if (lock)
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): LOCK prefix found twice", (size_t)code - out_length);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): LOCK prefix found twice", (size_t)code - out_length);
}
lock = true;
@ -150,7 +150,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
{
if (repne)
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): REPNE/REPNZ prefix found twice", (size_t)code - out_length);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): REPNE/REPNZ prefix found twice", (size_t)code - out_length);
}
repne = true;
@ -160,7 +160,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
{
if (repe)
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): REP/REPE/REPZ prefix found twice", (size_t)code - out_length);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): REP/REPE/REPZ prefix found twice", (size_t)code - out_length);
}
repe = true;
@ -176,7 +176,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
{
if (pg2)
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): 0x%02x (group 2 prefix) found after 0x%02x", (size_t)code - out_length, prefix, pg2);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): 0x%02x (group 2 prefix) found after 0x%02x", (size_t)code - out_length, prefix, pg2);
}
else
{
@ -189,7 +189,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
{
if (oso)
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): operand-size override prefix found twice", (size_t)code - out_length);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): operand-size override prefix found twice", (size_t)code - out_length);
}
oso = true;
@ -198,7 +198,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
case 0x67: // group 4
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): address-size override prefix found", (size_t)code - out_length, prefix);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): address-size override prefix found", (size_t)code - out_length, prefix);
out_op = X64OP_NONE;
out_reg = X64_NOT_SET;
out_size = 0;
@ -212,7 +212,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
{
if (rex)
{
LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): 0x%02x (REX prefix) found after 0x%02x", (size_t)code - out_length, prefix, rex);
LOG_ERROR(MEMORY, "decode_x64_reg_op(%016llxh): 0x%02x (REX prefix) found after 0x%02x", (size_t)code - out_length, prefix, rex);
}
else
{
@ -423,7 +423,7 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
}
}
LOG_WARNING(GENERAL, "decode_x64_reg_op(%016llxh): unsupported opcode found (%016llX%016llX)", (size_t)code - out_length, *(be_t<u64>*)(code - out_length), *(be_t<u64>*)(code - out_length + 8));
LOG_WARNING(MEMORY, "decode_x64_reg_op(%016llxh): unsupported opcode found (%016llX%016llX)", (size_t)code - out_length, *(be_t<u64>*)(code - out_length), *(be_t<u64>*)(code - out_length + 8));
out_op = X64OP_NONE;
out_reg = X64_NOT_SET;
out_size = 0;
@ -434,7 +434,9 @@ void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, siz
typedef CONTEXT x64_context;
#define X64REG(context, reg) (&(&context->Rax)[reg])
#define X64REG(context, reg) (&(&(context)->Rax)[reg])
#define XMMREG(context, reg) (reinterpret_cast<u128*>(&(&(context)->Xmm0)[reg]))
#define EFLAGS(context) ((context)->EFlags)
#else
@ -443,6 +445,8 @@ typedef ucontext_t x64_context;
#ifdef __APPLE__
#define X64REG(context, reg) (darwin_x64reg(context, reg))
#define XMMREG(context, reg) (reinterpret_cast<u128*>(&(context)->uc_mcontext->__fs.__fpu_xmm0[reg]))
#define EFLAGS(context) ((context)->uc_mcontext->__ss.__eflags)
uint64_t* darwin_x64reg(x64_context *context, int reg)
{
@ -498,7 +502,9 @@ static const reg_table_t reg_table[17] =
REG_R8, REG_R9, REG_R10, REG_R11, REG_R12, REG_R13, REG_R14, REG_R15, REG_RIP
};
#define X64REG(context, reg) (&context->uc_mcontext.gregs[reg_table[reg]])
#define X64REG(context, reg) (&(context)->uc_mcontext.gregs[reg_table[reg]])
#define XMMREG(context, reg) (reinterpret_cast<u128*>(&(context)->uc_mcontext.fpregs->_xmm[reg]))
#define EFLAGS(context) ((context)->uc_mcontext.eflags)
#endif // __APPLE__
@ -554,7 +560,7 @@ bool get_x64_reg_value(x64_context* context, x64_reg_t reg, size_t d_size, size_
return true;
}
LOG_ERROR(GENERAL, "get_x64_reg_value(): invalid arguments (reg=%d, d_size=%lld, i_size=%lld)", reg, d_size, i_size);
LOG_ERROR(MEMORY, "get_x64_reg_value(): invalid arguments (reg=%d, d_size=%lld, i_size=%lld)", reg, d_size, i_size);
return false;
}
@ -563,36 +569,113 @@ bool put_x64_reg_value(x64_context* context, x64_reg_t reg, size_t d_size, u64 v
// save x64 reg value (for load operations)
if (reg - X64R_RAX < 16)
{
// store the value into x64 register
*X64REG(context, reg - X64R_RAX) = (u32)value;
return true;
// save the value into x64 register
switch (d_size)
{
case 1: *X64REG(context, reg - X64R_RAX) = value & 0xff | *X64REG(context, reg - X64R_RAX) & 0xffffff00; return true;
case 2: *X64REG(context, reg - X64R_RAX) = value & 0xffff | *X64REG(context, reg - X64R_RAX) & 0xffff0000; return true;
case 4: *X64REG(context, reg - X64R_RAX) = value & 0xffffffff; return true;
case 8: *X64REG(context, reg - X64R_RAX) = value; return true;
}
}
LOG_ERROR(GENERAL, "put_x64_reg_value(): invalid destination (reg=%d, d_size=%lld, value=0x%llx)", reg, d_size, value);
LOG_ERROR(MEMORY, "put_x64_reg_value(): invalid destination (reg=%d, d_size=%lld, value=0x%llx)", reg, d_size, value);
return false;
}
void fix_x64_reg_op(x64_context* context, x64_op_t& op, x64_reg_t& reg, size_t& d_size, size_t& i_size)
bool set_x64_cmp_flags(x64_context* context, size_t d_size, u64 x, u64 y)
{
switch (d_size)
{
case 1: break;
case 2: break;
case 4: break;
case 8: break;
default: LOG_ERROR(MEMORY, "set_x64_cmp_flags(): invalid d_size (%lld)", d_size); return false;
}
const u64 sign = 1ull << (d_size * 8 - 1); // sign mask
const u64 diff = x - y;
const u64 summ = x + y;
if (((x & y) | ((x ^ y) & ~summ)) & sign)
{
EFLAGS(context) |= 0x1; // set CF
}
else
{
EFLAGS(context) &= ~0x1; // clear CF
}
if (x == y)
{
EFLAGS(context) |= 0x40; // set ZF
}
else
{
EFLAGS(context) &= ~0x40; // clear ZF
}
if (diff & sign)
{
EFLAGS(context) |= 0x80; // set SF
}
else
{
EFLAGS(context) &= ~0x80; // clear SF
}
if ((x ^ summ) & (y ^ summ) & sign)
{
EFLAGS(context) |= 0x800; // set OF
}
else
{
EFLAGS(context) &= ~0x800; // clear OF
}
const u8 p1 = (u8)diff ^ ((u8)diff >> 4);
const u8 p2 = p1 ^ (p1 >> 2);
const u8 p3 = p2 ^ (p2 >> 1);
if ((p3 & 1) == 0)
{
EFLAGS(context) |= 0x4; // set PF
}
else
{
EFLAGS(context) &= ~0x4; // clear PF
}
if (((x & y) | ((x ^ y) & ~summ)) & 0x8)
{
EFLAGS(context) |= 0x10; // set AF
}
else
{
EFLAGS(context) &= ~0x10; // clear AF
}
return true;
}
size_t get_x64_access_size(x64_context* context, x64_op_t op, x64_reg_t reg, size_t d_size, size_t i_size)
{
if (op == X64OP_MOVS && reg != X64_NOT_SET) // get "full" access size from RCX register
{
u64 counter;
if (!get_x64_reg_value(context, reg, 8, i_size, counter))
{
op = X64OP_NONE;
reg = X64_NOT_SET;
d_size = 0;
i_size = 0;
return;
return ~0ull;
}
d_size *= counter;
reg = X64_NOT_SET;
return;
return d_size * counter;
}
return d_size;
}
bool handle_access_violation(const u32 addr, bool is_writing, x64_context* context)
bool handle_access_violation(u32 addr, bool is_writing, x64_context* context)
{
auto code = (const u8*)RIP(context);
@ -603,37 +686,28 @@ bool handle_access_violation(const u32 addr, bool is_writing, x64_context* conte
// decode single x64 instruction that causes memory access
decode_x64_reg_op(code, op, reg, d_size, i_size);
fix_x64_reg_op(context, op, reg, d_size, i_size);
if (d_size + addr >= 0x100000000ull)
if ((d_size | d_size + addr) >= 0x100000000ull)
{
LOG_ERROR(GENERAL, "Invalid d_size (0x%llx)", d_size);
LOG_ERROR(MEMORY, "Invalid d_size (0x%llx)", d_size);
return false;
}
if (op == X64OP_CMPXCHG)
{
// detect whether this instruction can't actually modify memory to avoid breaking reservation;
// this may theoretically cause endless loop, but it shouldn't be a problem if only read_sync() generates such instruction
u64 cmp, exch;
if (!get_x64_reg_value(context, reg, d_size, i_size, cmp) || !get_x64_reg_value(context, X64R_RAX, d_size, i_size, exch))
{
return false;
}
// get length of data being accessed
size_t a_size = get_x64_access_size(context, op, reg, d_size, i_size);
if (cmp == exch)
{
// could also be emulated without attempt to write memory
is_writing = false;
}
if ((a_size | a_size + addr) >= 0x100000000ull)
{
LOG_ERROR(MEMORY, "Invalid a_size (0x%llx)", a_size);
return false;
}
// check if address is RawSPU MMIO register
if (addr - RAW_SPU_BASE_ADDR < (6 * RAW_SPU_OFFSET) && (addr % RAW_SPU_OFFSET) >= RAW_SPU_PROB_OFFSET)
{
if (d_size != 4 || !i_size)
if (a_size != 4 || !d_size || !i_size)
{
LOG_ERROR(GENERAL, "Invalid instruction (op=%d, reg=%d, d_size=%lld, i_size=%lld)", op, reg, d_size, i_size);
LOG_ERROR(MEMORY, "Invalid or unsupported instruction (op=%d, reg=%d, d_size=%lld, a_size=0x%llx, i_size=%lld)", op, reg, d_size, a_size, i_size);
return false;
}
@ -662,7 +736,7 @@ bool handle_access_violation(const u32 addr, bool is_writing, x64_context* conte
case X64OP_MOVS: // TODO
default:
{
LOG_ERROR(GENERAL, "Invalid operation (op=%d)", op);
LOG_ERROR(MEMORY, "Invalid or unsupported operation (op=%d, reg=%d, d_size=%lld, i_size=%lld)", op, reg, d_size, i_size);
return false;
}
}
@ -672,14 +746,189 @@ bool handle_access_violation(const u32 addr, bool is_writing, x64_context* conte
return true;
}
// check if fault is caused by reservation
if (vm::reservation_query(addr, (u32)d_size, is_writing))
if (op == X64OP_CMPXCHG)
{
return true;
// detect whether this instruction can't actually modify memory to avoid breaking reservation;
// this may theoretically cause endless loop, but it shouldn't be a problem if only read_sync() generates such instruction
u64 cmp, exch;
if (!get_x64_reg_value(context, reg, d_size, i_size, cmp) || !get_x64_reg_value(context, X64R_RAX, d_size, i_size, exch))
{
return false;
}
if (cmp == exch)
{
// this will skip reservation bound check
a_size = 0;
}
}
// check if fault is caused by the reservation
return vm::reservation_query(addr, (u32)a_size, is_writing, [&]() -> bool
{
// write memory using "privileged" access to avoid breaking reservation
if (!d_size || !i_size)
{
LOG_ERROR(MEMORY, "Invalid or unsupported instruction (op=%d, reg=%d, d_size=%lld, a_size=0x%llx, i_size=%lld)", op, reg, d_size, a_size, i_size);
return false;
}
switch (op)
{
case X64OP_STORE:
{
if (d_size == 16)
{
if (reg - X64R_XMM0 >= 16)
{
LOG_ERROR(MEMORY, "X64OP_STORE: d_size=16, reg=%d", reg);
return false;
}
memcpy(vm::get_priv_ptr(addr), XMMREG(context, reg - X64R_XMM0), 16);
break;
}
if (d_size > 8)
{
LOG_ERROR(MEMORY, "X64OP_STORE: d_size=%lld", d_size);
return false;
}
u64 reg_value;
if (!get_x64_reg_value(context, reg, d_size, i_size, reg_value))
{
return false;
}
memcpy(vm::get_priv_ptr(addr), &reg_value, d_size);
break;
}
case X64OP_MOVS:
{
if (d_size > 8)
{
LOG_ERROR(MEMORY, "X64OP_MOVS: d_size=%lld", d_size);
return false;
}
if (vm::get_ptr(addr) != (void*)RDI(context))
{
LOG_ERROR(MEMORY, "X64OP_MOVS error: rdi=0x%llx, addr=0x%x", RDI(context), addr);
return false;
}
u32 a_addr = addr;
while (a_addr >> 12 == addr >> 12)
{
u64 value;
// copy data
memcpy(&value, (void*)RSI(context), d_size);
memcpy(vm::get_priv_ptr(a_addr), &value, d_size);
// shift pointers
if (EFLAGS(context) & 0x400 /* direction flag */)
{
// for reversed direction, addr argument should be calculated in different way
LOG_ERROR(MEMORY, "X64OP_MOVS TODO: reversed direction");
return false;
//RSI(context) -= d_size;
//RDI(context) -= d_size;
//a_addr -= (u32)d_size;
}
else
{
RSI(context) += d_size;
RDI(context) += d_size;
a_addr += (u32)d_size;
}
// decrement counter
if (reg == X64_NOT_SET || !--RCX(context))
{
break;
}
}
if (reg == X64_NOT_SET || !RCX(context))
{
break;
}
// don't skip partially processed instruction
return true;
}
case X64OP_XCHG:
{
if (d_size != 1 && d_size != 2 && d_size != 4 && d_size != 8)
{
LOG_ERROR(MEMORY, "X64OP_XCHG: d_size=%lld", d_size);
return false;
}
u64 reg_value;
if (!get_x64_reg_value(context, reg, d_size, i_size, reg_value))
{
return false;
}
switch (d_size)
{
case 1: reg_value = vm::get_priv_ref<atomic_le_t<u8>>(addr).exchange((u8)reg_value); break;
case 2: reg_value = vm::get_priv_ref<atomic_le_t<u16>>(addr).exchange((u16)reg_value); break;
case 4: reg_value = vm::get_priv_ref<atomic_le_t<u32>>(addr).exchange((u32)reg_value); break;
case 8: reg_value = vm::get_priv_ref<atomic_le_t<u64>>(addr).exchange((u64)reg_value); break;
}
if (!put_x64_reg_value(context, reg, d_size, reg_value))
{
return false;
}
break;
}
case X64OP_CMPXCHG:
{
if (d_size != 1 && d_size != 2 && d_size != 4 && d_size != 8)
{
LOG_ERROR(MEMORY, "X64OP_CMPXCHG: d_size=%lld", d_size);
return false;
}
u64 reg_value, old_value, cmp_value;
if (!get_x64_reg_value(context, reg, d_size, i_size, reg_value) || !get_x64_reg_value(context, X64R_RAX, d_size, i_size, cmp_value))
{
return false;
}
switch (d_size)
{
case 1: old_value = vm::get_priv_ref<atomic_le_t<u8>>(addr).compare_and_swap((u8)cmp_value, (u8)reg_value); break;
case 2: old_value = vm::get_priv_ref<atomic_le_t<u16>>(addr).compare_and_swap((u16)cmp_value, (u16)reg_value); break;
case 4: old_value = vm::get_priv_ref<atomic_le_t<u32>>(addr).compare_and_swap((u32)cmp_value, (u32)reg_value); break;
case 8: old_value = vm::get_priv_ref<atomic_le_t<u64>>(addr).compare_and_swap((u64)cmp_value, (u64)reg_value); break;
}
if (!put_x64_reg_value(context, X64R_RAX, d_size, old_value) || !set_x64_cmp_flags(context, d_size, cmp_value, old_value))
{
return false;
}
break;
}
default:
{
LOG_ERROR(MEMORY, "Invalid or unsupported operation (op=%d, reg=%d, d_size=%lld, a_size=0x%llx, i_size=%lld)", op, reg, d_size, a_size, i_size);
return false;
}
}
// skip processed instruction
RIP(context) += i_size;
return true;
});
// TODO: allow recovering from a page fault as a feature of PS3 virtual memory
return false;
}
#ifdef _WIN32

View File

@ -1,5 +1,11 @@
#pragma once
#undef InterlockedExchange
#undef InterlockedCompareExchange
#undef InterlockedOr
#undef InterlockedAnd
#undef InterlockedXor
template<typename T, size_t size = sizeof(T)>
struct _to_atomic
{

View File

@ -271,7 +271,7 @@ namespace vm
return true;
}
bool reservation_query(u32 addr, u32 size, bool is_writing)
bool reservation_query(u32 addr, u32 size, bool is_writing, std::function<bool()> callback)
{
std::lock_guard<reservation_mutex_t> lock(g_reservation_mutex);
@ -280,31 +280,17 @@ namespace vm
return false;
}
if (is_writing)
// check if current reservation and address may overlap
if (g_reservation_addr >> 12 == addr >> 12 && is_writing)
{
assert(size);
if (addr + size - 1 >= g_reservation_addr && g_reservation_addr + g_reservation_size - 1 >= addr)
if (size && addr + size - 1 >= g_reservation_addr && g_reservation_addr + g_reservation_size - 1 >= addr)
{
// break the reservation if writing access and reservation overlap
// break the reservation if overlap
_reservation_break(addr);
}
else
{
// full-size check (isn't accurate enough)
if (!check_addr(addr, size))
{
return false;
}
// assume that the same memory page is accessed (isn't accurate enough)
if (g_reservation_addr >> 12 != addr >> 12)
{
return false;
}
// write memory using "privileged" access to avoid breaking reservation
return false;
return callback(); //? true : _reservation_break(addr), true;
}
}

View File

@ -39,7 +39,7 @@ namespace vm
// attempt to atomically update reserved memory
bool reservation_update(u32 addr, const void* data, u32 size);
// for internal use
bool reservation_query(u32 addr, u32 size, bool is_writing);
bool reservation_query(u32 addr, u32 size, bool is_writing, std::function<bool()> callback);
// for internal use
void reservation_free();
// perform complete operation