ShuriZma
/
rpcs3
mirror of https://github.com/RPCS3/rpcs3.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

PPU LLVM: Implement SELF precompilation 

Do not use PS3 memory for precompilation.
This commit is contained in:
Eladash 2023-06-25 15:53:42 +03:00 committed by Ivan
parent 7062ead4fd
commit 554b27a82a
11 changed files with 459 additions and 214 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
Utilities/bin_patch.cpp
View File

@ -805,7 +805,7 @@ void unmap_vm_area(std::shared_ptr<vm::block_t>& ptr)
}
// Returns old 'applied' size
static usz apply_modification(std::basic_string<u32>& applied, patch_engine::patch_info& patch, u8* dst, u32 filesz, u32 min_addr)
static usz apply_modification(std::basic_string<u32>& applied, patch_engine::patch_info& patch, std::function<u8*(u32)> mem_translate, u32 filesz, u32 min_addr)
{
const usz old_applied_size = applied.size();
@ -846,8 +846,8 @@ static usz apply_modification(std::basic_string<u32>& applied, patch_engine::pat
{
if (p.type != patch_type::alloc) continue;
// Do not allow null address or if dst is not a VM ptr
if (const u32 alloc_at = vm::try_get_addr(dst + (p.offset & -4096)).first; alloc_at >> 16)
// Do not allow null address or if resultant ptr is not a VM ptr
if (const u32 alloc_at = vm::try_get_addr(mem_translate(p.offset & -4096)).first; alloc_at >> 16)
{
const u32 alloc_size = utils::align(static_cast<u32>(p.value.long_value) + alloc_at % 4096, 4096);
@ -934,7 +934,13 @@ static usz apply_modification(std::basic_string<u32>& applied, patch_engine::pat
offset -= min_addr;
auto ptr = dst + offset;
auto ptr = mem_translate(offset);
if (!ptr)
{
// Memory translation failed
continue;
}
if (relocate_instructions_at)
{
@ -960,7 +966,7 @@ static usz apply_modification(std::basic_string<u32>& applied, patch_engine::pat
}
case patch_type::code_alloc:
{
const u32 out_branch = vm::try_get_addr(dst + (offset & -4)).first;
const u32 out_branch = vm::try_get_addr(mem_translate(offset & -4)).first;
// Allow only if points to a PPU executable instruction
if (out_branch < 0x10000 || out_branch >= 0x4000'0000 || !vm::check_addr<4>(out_branch, vm::page_executable))
@ -1044,7 +1050,7 @@ static usz apply_modification(std::basic_string<u32>& applied, patch_engine::pat
case patch_type::jump:
case patch_type::jump_link:
{
const u32 out_branch = vm::try_get_addr(dst + (offset & -4)).first;
const u32 out_branch = vm::try_get_addr(mem_translate(offset & -4)).first;
const u32 dest = static_cast<u32>(p.value.long_value);
// Allow only if points to a PPU executable instruction
@ -1060,7 +1066,7 @@ static usz apply_modification(std::basic_string<u32>& applied, patch_engine::pat
{
const std::string& str = p.original_value;
const u32 out_branch = vm::try_get_addr(dst + (offset & -4)).first;
const u32 out_branch = vm::try_get_addr(mem_translate(offset & -4)).first;
const usz sep_pos = str.find_first_of(':');
// Must contain only a single ':' or none
@ -1251,7 +1257,7 @@ static usz apply_modification(std::basic_string<u32>& applied, patch_engine::pat
return old_applied_size;
}
std::basic_string<u32> patch_engine::apply(const std::string& name, u8* dst, u32 filesz, u32 min_addr)
std::basic_string<u32> patch_engine::apply(const std::string& name, std::function<u8*(u32)> mem_translate, u32 filesz, u32 min_addr)
{
if (!m_map.contains(name))
{
@ -1392,7 +1398,7 @@ std::basic_string<u32> patch_engine::apply(const std::string& name, u8* dst, u32
{
if (patch)
{
const usz old_size = apply_modification(applied_total, *patch, dst, filesz, min_addr);
const usz old_size = apply_modification(applied_total, *patch, mem_translate, filesz, min_addr);
if (applied_total.size() != old_size)
{

3
Utilities/bin_patch.h
View File

@ -3,6 +3,7 @@
#include <vector>
#include <string>
#include <unordered_map>
#include <functional>
#include "util/types.hpp"
#include "util/yaml.hpp"
@ -212,7 +213,7 @@ public:
void append_title_patches(const std::string& title_id);
// Apply patch (returns the number of entries applied)
std::basic_string<u32> apply(const std::string& name, u8* dst, u32 filesz = -1, u32 min_addr = 0);
std::basic_string<u32> apply(const std::string& name, std::function<u8*(u32)> mem_translate, u32 filesz = -1, u32 min_addr = 0);
// Deallocate memory used by patches
void unload(const std::string& name);

154
rpcs3/Emu/Cell/PPUAnalyser.cpp
View File

@ -78,7 +78,7 @@ void ppu_module::validate(u32 reloc)
if (size && size != funcs[index].size)
{
if (size + 4 != funcs[index].size || vm::read32(addr + size) != ppu_instructions::NOP())
if (size + 4 != funcs[index].size || *ensure(get_ptr<u32>(addr + size)) != ppu_instructions::NOP())
{
ppu_validator.error("%s.yml : function size mismatch at 0x%x(size=0x%x) (0x%x, 0x%x)", path, found, funcs[index].size, addr, size);
}
@ -112,9 +112,9 @@ void ppu_module::validate(u32 reloc)
}
}
static u32 ppu_test(const vm::cptr<u32> ptr, vm::cptr<void> fend, ppu_pattern_array pat)
static u32 ppu_test(const be_t<u32>* ptr, const void* fend, ppu_pattern_array pat)
{
vm::cptr<u32> cur = ptr;
const be_t<u32>* cur = ptr;
for (auto& p : pat)
{
@ -141,10 +141,10 @@ static u32 ppu_test(const vm::cptr<u32> ptr, vm::cptr<void> fend, ppu_pattern_ar
cur++;
}
return cur.addr() - ptr.addr();
return (cur - ptr) * sizeof(*ptr);
}
static u32 ppu_test(vm::cptr<u32> ptr, vm::cptr<void> fend, ppu_pattern_matrix pats)
static u32 ppu_test(const be_t<u32>* ptr, const void* fend, ppu_pattern_matrix pats)
{
for (auto pat : pats)
{
@ -585,6 +585,14 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
return func;
};
static const auto advance = [](auto& _ptr, auto& ptr, u32 count)
{
const auto old_ptr = ptr;
_ptr += count;
ptr += count;
return old_ptr;
};
// Register new TOC and find basic set of functions
auto add_toc = [&](u32 toc)
{
@ -596,16 +604,24 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
// Grope for OPD section (TODO: optimization, better constraints)
for (const auto& seg : segs)
{
if (!seg.addr) continue;
if (seg.size < 8) continue;
for (vm::cptr<u32> ptr = vm::cast(seg.addr); ptr.addr() < seg.addr + seg.size; ptr++)
const vm::cptr<void> seg_end = vm::cast(seg.addr + seg.size - 8);
vm::cptr<u32> _ptr = vm::cast(seg.addr);
auto ptr = get_ptr<u32>(_ptr);
for (; _ptr <= seg_end;)
{
if (ptr[0] >= start && ptr[0] < end && ptr[0] % 4 == 0 && ptr[1] == toc)
if (ptr[1] == toc && ptr[0] >= start && ptr[0] < end && ptr[0] % 4 == 0)
{
// New function
ppu_log.trace("OPD*: [0x%x] 0x%x (TOC=0x%x)", ptr, ptr[0], ptr[1]);
add_func(*ptr, addr_heap.count(ptr.addr()) ? toc : 0, 0);
ptr++;
ppu_log.trace("OPD*: [0x%x] 0x%x (TOC=0x%x)", _ptr, ptr[0], ptr[1]);
add_func(*ptr, addr_heap.count(_ptr.addr()) ? toc : 0, 0);
advance(_ptr, ptr, 2);
}
else
{
advance(_ptr, ptr, 1);
}
}
}
@ -621,9 +637,13 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
// Find references indiscriminately
for (const auto& seg : segs)
{
if (!seg.addr) continue;
if (seg.size < 4) continue;
for (vm::cptr<u32> ptr = vm::cast(seg.addr); ptr.addr() < seg.addr + seg.size; ptr++)
vm::cptr<u32> _ptr = vm::cast(seg.addr);
const vm::cptr<void> seg_end = vm::cast(seg.addr + seg.size - 4);
auto ptr = get_ptr<u32>(_ptr);
for (vm::cptr<u32> _ptr = vm::cast(seg.addr); _ptr <= seg_end; advance(_ptr, ptr, 1))
{
const u32 value = *ptr;
@ -651,15 +671,17 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
vm::cptr<void> sec_end = vm::cast(sec.addr + sec.size);
// Probe
for (vm::cptr<u32> ptr = vm::cast(sec.addr); ptr < sec_end; ptr += 2)
for (vm::cptr<u32> _ptr = vm::cast(sec.addr); _ptr < sec_end; _ptr += 2)
{
if (ptr + 6 <= sec_end && !ptr[0] && !ptr[2] && ptr[1] == ptr[4] && ptr[3] == ptr[5])
auto ptr = get_ptr<u32>(_ptr);
if (_ptr + 6 <= sec_end && !ptr[0] && !ptr[2] && ptr[1] == ptr[4] && ptr[3] == ptr[5])
{
// Special OPD format case (some homebrews)
ptr += 4;
advance(_ptr, ptr, 4);
}
if (ptr + 2 > sec_end)
if (_ptr + 2 > sec_end)
{
sec_end.set(0);
break;
@ -673,7 +695,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
//const u32 _toc_end = _toc + 0x8000;
// TODO: improve TOC constraints
if (_toc % 4 || !vm::check_addr(_toc) || _toc >= 0x40000000 || (_toc >= start && _toc < end))
if (_toc % 4 || !get_ptr<u32>(_toc) || _toc >= 0x40000000 || (_toc >= start && _toc < end))
{
sec_end.set(0);
break;
@ -689,18 +711,20 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
if (sec_end) ppu_log.notice("Reading OPD section at 0x%x...", sec.addr);
// Mine
for (vm::cptr<u32> ptr = vm::cast(sec.addr); ptr < sec_end; ptr += 2)
for (vm::cptr<u32> _ptr = vm::cast(sec.addr); _ptr < sec_end; _ptr += 2)
{
auto ptr = get_ptr<u32>(_ptr);
// Special case: see "Probe"
if (!ptr[0]) ptr += 4;
if (!ptr[0]) advance(_ptr, ptr, 4);
// Add function and TOC
const u32 addr = ptr[0];
const u32 toc = ptr[1];
ppu_log.trace("OPD: [0x%x] 0x%x (TOC=0x%x)", ptr, addr, toc);
ppu_log.trace("OPD: [0x%x] 0x%x (TOC=0x%x)", _ptr, addr, toc);
TOCs.emplace(toc);
auto& func = add_func(addr, addr_heap.count(ptr.addr()) ? toc : 0, 0);
auto& func = add_func(addr, addr_heap.count(_ptr.addr()) ? toc : 0, 0);
func.attr += ppu_attr::known_addr;
known_functions.emplace(addr);
}
@ -709,7 +733,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
// Register TOC from entry point
if (entry && !lib_toc)
{
lib_toc = vm::read32(entry) ? vm::read32(entry + 4) : vm::read32(entry + 20);
lib_toc = *ensure(get_ptr<u32>(entry)) ? *ensure(get_ptr<u32>(entry + 4)) : *ensure(get_ptr<u32>(entry + 20));
}
// Secondary attempt
@ -733,23 +757,25 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
vm::cptr<void> sec_end = vm::cast(sec.addr + sec.size);
// Probe
for (vm::cptr<u32> ptr = vm::cast(sec.addr); ptr < sec_end;)
for (vm::cptr<u32> _ptr = vm::cast(sec.addr); _ptr < sec_end;)
{
if (!ptr.aligned() || ptr.addr() < sec.addr || ptr >= sec_end)
if (!_ptr.aligned() || _ptr.addr() < sec.addr || _ptr >= sec_end)
{
sec_end.set(0);
break;
}
const auto ptr = get_ptr<u32>(_ptr);
const u32 size = ptr[0] + 4;
if (size == 4 && ptr + 1 == sec_end)
if (size == 4 && _ptr + 1 == sec_end)
{
// Null terminator
break;
}
if (size % 4 || size < 0x10 || ptr + size / 4 > sec_end)
if (size % 4 || size < 0x10 || _ptr + size / 4 > sec_end)
{
sec_end.set(0);
break;
@ -757,7 +783,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
if (ptr[1])
{
const u32 cie_off = ptr.addr() - ptr[1] + 4;
const u32 cie_off = _ptr.addr() - ptr[1] + 4;
if (cie_off % 4 || cie_off < sec.addr || cie_off >= sec_end.addr())
{
@ -766,14 +792,16 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
}
}
ptr = vm::cast(ptr.addr() + size);
_ptr = vm::cast(_ptr.addr() + size);
}
if (sec_end && sec.size > 4) ppu_log.notice("Reading .eh_frame section at 0x%x...", sec.addr);
// Mine
for (vm::cptr<u32> ptr = vm::cast(sec.addr); ptr < sec_end; ptr = vm::cast(ptr.addr() + ptr[0] + 4))
for (vm::cptr<u32> _ptr = vm::cast(sec.addr); _ptr < sec_end; _ptr = vm::cast(_ptr.addr() + *get_ptr<u32>(_ptr) + 4))
{
const auto ptr = get_ptr<u32>(_ptr);
if (ptr[0] == 0u)
{
// Null terminator
@ -788,7 +816,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
else
{
// Get associated CIE (currently unused)
const vm::cptr<u32> cie = vm::cast(ptr.addr() - ptr[1] + 4);
const vm::cptr<u32> cie = vm::cast(_ptr.addr() - ptr[1] + 4);
u32 addr = 0;
u32 size = 0;
@ -817,7 +845,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
// TODO: absolute/relative offset (approximation)
if (addr > 0xc0000000)
{
addr += ptr.addr() + 8;
addr += _ptr.addr() + 8;
}
ppu_log.trace(".eh_frame: [0x%x] FDE 0x%x (cie=*0x%x, addr=0x%x, size=0x%x)", ptr, ptr[0], cie, addr, size);
@ -875,15 +903,16 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
if (func.blocks.empty())
{
// Special function analysis
const vm::cptr<u32> ptr = vm::cast(func.addr);
const vm::cptr<u32> _ptr = vm::cast(func.addr);
const vm::cptr<void> fend = vm::cast(end);
const auto ptr = get_ptr<u32>(_ptr);
using namespace ppu_instructions;
if (ptr + 1 <= fend && (ptr[0] & 0xfc000001) == B({}, {}))
if (_ptr + 1 <= fend && (ptr[0] & 0xfc000001) == B({}, {}))
{
// Simple trampoline
const u32 target = (ptr[0] & 0x2 ? 0 : ptr.addr()) + ppu_opcode_t{ptr[0]}.bt24;
const u32 target = (ptr[0] & 0x2 ? 0 : _ptr.addr()) + ppu_opcode_t{ptr[0]}.bt24;
if (target == func.addr)
{
@ -913,7 +942,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
}
}
if (ptr + 0x4 <= fend &&
if (_ptr + 0x4 <= fend &&
(ptr[0] & 0xffff0000) == LIS(r11, 0) &&
(ptr[1] & 0xffff0000) == ADDI(r11, r11, 0) &&
ptr[2] == MTCTR(r11) &&
@ -941,7 +970,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
}
}
if (ptr + 0x7 <= fend &&
if (_ptr + 0x7 <= fend &&
ptr[0] == STD(r2, r1, 0x28) &&
(ptr[1] & 0xffff0000) == ADDIS(r12, r2, {}) &&
(ptr[2] & 0xffff0000) == LWZ(r11, r12, {}) &&
@ -957,9 +986,10 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
func.attr += ppu_attr::known_size;
// Look for another imports to fill gaps (hack)
auto p2 = ptr + 7;
auto _p2 = _ptr + 7;
auto p2 = get_ptr<u32>(_p2);
while (p2 + 0x7 <= fend &&
while (_p2 + 0x7 <= fend &&
p2[0] == STD(r2, r1, 0x28) &&
(p2[1] & 0xffff0000) == ADDIS(r12, r2, {}) &&
(p2[2] & 0xffff0000) == LWZ(r11, r12, {}) &&
@ -968,18 +998,18 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
p2[5] == MTCTR(r11) &&
p2[6] == BCTR())
{
auto& next = add_func(p2.addr(), -1, func.addr);
auto& next = add_func(_p2.addr(), -1, func.addr);
next.size = 0x1C;
next.blocks.emplace(next.addr, next.size);
next.attr += ppu_attr::known_addr;
next.attr += ppu_attr::known_size;
p2 += 7;
advance(_p2, p2, 7);
}
continue;
}
if (ptr + 0x7 <= fend &&
if (_ptr + 0x7 <= fend &&
ptr[0] == STD(r2, r1, 0x28) &&
(ptr[1] & 0xffff0000) == ADDIS(r2, r2, {}) &&
(ptr[2] & 0xffff0000) == ADDI(r2, r2, {}) &&
@ -1029,7 +1059,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
}
}
if (ptr + 4 <= fend &&
if (_ptr + 4 <= fend &&
ptr[0] == STD(r2, r1, 0x28) &&
(ptr[1] & 0xffff0000) == ADDIS(r2, r2, {}) &&
(ptr[2] & 0xffff0000) == ADDI(r2, r2, {}) &&
@ -1037,7 +1067,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
{
// Trampoline with TOC
const u32 toc_add = (ptr[1] << 16) + s16(ptr[2]);
const u32 target = (ptr[3] & 0x2 ? 0 : (ptr + 3).addr()) + ppu_opcode_t{ptr[3]}.bt24;
const u32 target = (ptr[3] & 0x2 ? 0 : (_ptr + 3).addr()) + ppu_opcode_t{ptr[3]}.bt24;
if (target >= start && target < end)
{
@ -1076,7 +1106,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
}
}
if (ptr + 8 <= fend &&
if (_ptr + 8 <= fend &&
(ptr[0] & 0xffff0000) == LI(r12, 0) &&
(ptr[1] & 0xffff0000) == ORIS(r12, r12, 0) &&
(ptr[2] & 0xffff0000) == LWZ(r12, r12, 0) &&
@ -1095,9 +1125,10 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
func.attr += ppu_attr::known_size;
// Look for another imports to fill gaps (hack)
auto p2 = ptr + 8;
auto _p2 = _ptr + 8;
auto p2 = get_ptr<u32>(_p2);
while (p2 + 8 <= fend &&
while (_p2 + 8 <= fend &&
(p2[0] & 0xffff0000) == LI(r12, 0) &&
(p2[1] & 0xffff0000) == ORIS(r12, r12, 0) &&
(p2[2] & 0xffff0000) == LWZ(r12, r12, 0) &&
@ -1107,19 +1138,19 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
p2[6] == MTCTR(r0) &&
p2[7] == BCTR())
{
auto& next = add_func(p2.addr(), -1, func.addr);
auto& next = add_func(_p2.addr(), -1, func.addr);
next.size = 0x20;
next.blocks.emplace(next.addr, next.size);
next.attr += ppu_attr::known_addr;
next.attr += ppu_attr::known_size;
p2 += 8;
advance(_p2, p2, 8);
known_functions.emplace(next.addr);
}
continue;
}
if (ptr + 3 <= fend &&
if (_ptr + 3 <= fend &&
ptr[0] == 0x7c0004acu &&
ptr[1] == 0x00000000u &&
ptr[2] == BLR())
@ -1131,7 +1162,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
continue;
}
if (const u32 len = ppu_test(ptr, fend, ppu_patterns::abort))
if (const u32 len = ppu_test(ptr, get_ptr<void>(fend), ppu_patterns::abort))
{
// Function "abort"
ppu_log.notice("Function [0x%x]: 'abort'", func.addr);
@ -1199,10 +1230,13 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
{
auto& block = block_queue[j].get();
for (vm::cptr<u32> _ptr = vm::cast(block.first); _ptr.addr() < func_end;)
vm::cptr<u32> _ptr = vm::cast(block.first);
auto ptr = ensure(get_ptr<u32>(_ptr));
for (; _ptr.addr() < func_end;)
{
const u32 iaddr = _ptr.addr();
const ppu_opcode_t op{*_ptr++};
const ppu_opcode_t op{*advance(_ptr, ptr, 1)};
const ppu_itype::type type = s_ppu_itype.decode(op.opcode);
if (type == ppu_itype::UNK)
@ -1276,9 +1310,9 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
const u32 jt_addr = _ptr.addr();
const u32 jt_end = func_end;
for (; _ptr.addr() < jt_end; _ptr++)
for (; _ptr.addr() < jt_end; advance(_ptr, ptr, 1))
{
const u32 addr = jt_addr + *_ptr;
const u32 addr = jt_addr + *ptr;
if (addr == jt_addr)
{
@ -1331,7 +1365,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
block.second = _ptr.addr() - block.first;
break;
}
else if (type == ppu_itype::STDU && func.attr & ppu_attr::no_size && (op.opcode == *_ptr || *_ptr == ppu_instructions::BLR()))
else if (type == ppu_itype::STDU && func.attr & ppu_attr::no_size && (op.opcode == *ptr || *ptr == ppu_instructions::BLR()))
{
// Hack
ppu_log.success("[0x%x] Instruction repetition: 0x%08x", iaddr, op.opcode);
@ -1391,7 +1425,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
for (vm::cptr<u32> _ptr = vm::cast(block.first); _ptr.addr() < block.first + block.second;)
{
const u32 iaddr = _ptr.addr();
const ppu_opcode_t op{*_ptr++};
const ppu_opcode_t op{*ensure(get_ptr<u32>(_ptr++))};
const ppu_itype::type type = s_ppu_itype.decode(op.opcode);
if (type == ppu_itype::B || type == ppu_itype::BC)
@ -1465,7 +1499,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
for (vm::cptr<u32> _ptr = vm::cast(start); _ptr.addr() < next;)
{
const u32 addr = _ptr.addr();
const ppu_opcode_t op{*_ptr++};
const ppu_opcode_t op{*ensure(get_ptr<u32>(_ptr++))};
const ppu_itype::type type = s_ppu_itype.decode(op.opcode);
if (type == ppu_itype::UNK)
@ -1607,7 +1641,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
continue;
}
const u32 target = vm::_ref<u32>(rel.addr);
const u32 target = *ensure(get_ptr<u32>(rel.addr));
if (target % 4 || target < start || target >= end)
{
@ -1684,7 +1718,7 @@ bool ppu_module::analyse(u32 lib_toc, u32 entry, const u32 sec_end, const std::b
for (; i_pos < lim; i_pos += 4)
{
const u32 opc = vm::_ref<u32>(i_pos);
const u32 opc = *ensure(get_ptr<u32>(i_pos));
switch (auto type = s_ppu_itype.decode(opc))
{

39
rpcs3/Emu/Cell/PPUAnalyser.h
View File

@ -3,8 +3,10 @@
#include <string>
#include <map>
#include <set>
#include <deque>
#include "util/types.hpp"
#include "util/endian.hpp"
#include "util/to_endian.hpp"
#include "Utilities/bit_set.h"
#include "PPUOpcodes.h"
@ -65,6 +67,7 @@ struct ppu_segment
u32 type;
u32 flags;
u32 filesz;
void* ptr{};
};
// PPU Module Information
@ -89,6 +92,8 @@ struct ppu_module
std::vector<ppu_segment> segs{};
std::vector<ppu_segment> secs{};
std::vector<ppu_function> funcs{};
std::deque<std::shared_ptr<void>> allocations;
std::map<u32, u32> addr_to_seg_index;
// Copy info without functions
void copy_part(const ppu_module& info)
@ -99,10 +104,44 @@ struct ppu_module
relocs = info.relocs;
segs = info.segs;
secs = info.secs;
allocations = info.allocations;
addr_to_seg_index = info.addr_to_seg_index;
}
bool analyse(u32 lib_toc, u32 entry, u32 end, const std::basic_string<u32>& applied, std::function<bool()> check_aborted = {});
void validate(u32 reloc);
template <typename T>
to_be_t<T>* get_ptr(u32 addr) const
{
auto it = addr_to_seg_index.upper_bound(addr);
if (it == addr_to_seg_index.begin())
{
return nullptr;
}
it--;
const auto& seg = segs[it->second];
const u32 seg_size = seg.size;
const u32 seg_addr = seg.addr;
constexpr usz size_element = std::is_void_v<T> ? 0 : sizeof(std::conditional_t<std::is_void_v<T>, char, T>);
if (seg_size >= std::max<usz>(size_element, 1) && addr <= seg_addr + seg_size - size_element)
{
return reinterpret_cast<to_be_t<T>*>(static_cast<u8*>(seg.ptr) + (addr - seg_addr));
}
return nullptr;
}
template <typename T, typename U> requires requires (const U& obj) { +obj.addr() * 0; }
to_be_t<T>* get_ptr(U&& addr) const
{
return get_ptr<T>(addr.addr());
}
};
struct main_ppu_module : public ppu_module

248
rpcs3/Emu/Cell/PPUModule.cpp
View File

@ -754,7 +754,7 @@ static auto ppu_load_exports(ppu_linkage_info* link, u32 exports_start, u32 expo
if (lib.num_tlsvar)
{
ppu_loader.fatal("Unexpected num_tlsvar (%u)!", lib.num_tlsvar);
ppu_loader.error("Unexpected num_tlsvar (%u)!", lib.num_tlsvar);
}
const bool should_load = ppu_register_library_lock(module_name, true);
@ -889,7 +889,7 @@ static auto ppu_load_imports(std::vector<ppu_reloc>& relocs, ppu_linkage_info* l
if (lib.num_tlsvar)
{
ppu_loader.fatal("Unexpected num_tlsvar (%u)!", lib.num_tlsvar);
ppu_loader.error("Unexpected num_tlsvar (%u)!", lib.num_tlsvar);
}
// Static module
@ -1056,13 +1056,18 @@ void init_ppu_functions(utils::serial* ar, bool full = false)
}
}
static void ppu_check_patch_spu_images(const ppu_segment& seg)
static void ppu_check_patch_spu_images(const ppu_module& mod, const ppu_segment& seg)
{
const std::string_view seg_view{vm::get_super_ptr<char>(seg.addr), seg.size};
if (!seg.size)
{
return;
}
const std::string_view seg_view{ensure(mod.get_ptr<char>(seg.addr)), seg.size};
for (usz i = seg_view.find("\177ELF"); i < seg.size; i = seg_view.find("\177ELF", i + 4))
{
const auto elf_header = vm::get_super_ptr<u8>(seg.addr + i);
const auto elf_header = ensure(mod.get_ptr<u8>(seg.addr + i));
// Try to load SPU image
const spu_exec_object obj(fs::file(elf_header, seg.size - i));
@ -1103,7 +1108,7 @@ static void ppu_check_patch_spu_images(const ppu_segment& seg)
sha1_update(&sha2, (elf_header + prog.p_offset), prog.p_filesz);
// We assume that the string SPUNAME exists 0x14 bytes into the NOTE segment
name = reinterpret_cast<char*>(elf_header + prog.p_offset + 0x14);
name = ensure(mod.get_ptr<const char>(seg.addr + i + prog.p_offset + 0x14));
if (!name.empty())
{
@ -1137,12 +1142,12 @@ static void ppu_check_patch_spu_images(const ppu_segment& seg)
for (const auto& prog : obj.progs)
{
// Apply the patch
applied += g_fxo->get<patch_engine>().apply(hash, (elf_header + prog.p_offset), prog.p_filesz, prog.p_vaddr);
applied += g_fxo->get<patch_engine>().apply(hash, [&](u32 addr) { return addr + elf_header + prog.p_offset; }, prog.p_filesz, prog.p_vaddr);
if (!Emu.GetTitleID().empty())
{
// Alternative patch
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, (elf_header + prog.p_offset), prog.p_filesz, prog.p_vaddr);
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, [&](u32 addr) { return addr + elf_header + prog.p_offset; }, prog.p_filesz, prog.p_vaddr);
}
}
@ -1248,7 +1253,7 @@ const char* get_prx_name_by_cia(u32 addr)
return nullptr;
}
std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::string& path, s64 file_offset, utils::serial* ar)
std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, bool virtual_load, const std::string& path, s64 file_offset, utils::serial* ar)
{
if (elf != elf_error::ok)
{
@ -1256,7 +1261,7 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
}
// Create new PRX object
const auto prx = !ar ? idm::make_ptr<lv2_obj, lv2_prx>() : std::make_shared<lv2_prx>();
const auto prx = !ar && !virtual_load ? idm::make_ptr<lv2_obj, lv2_prx>() : std::make_shared<lv2_prx>();
// Access linkage information object
auto& link = g_fxo->get<ppu_linkage_info>();
@ -1271,6 +1276,9 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
u32 end = 0;
u32 toc = 0;
// 0x100000: Workaround for analyser glitches
u32 allocating_address = 0x100000;
for (const auto& prog : elf.progs)
{
ppu_loader.notice("** Segment: p_type=0x%x, p_vaddr=0x%llx, p_filesz=0x%llx, p_memsz=0x%llx, flags=0x%x", prog.p_type, prog.p_vaddr, prog.p_filesz, prog.p_memsz, prog.p_flags);
@ -1295,15 +1303,42 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
// Alloc segment memory
// Or use saved address
const u32 addr = !ar ? vm::alloc(mem_size, vm::main) : ar->operator u32();
u32 addr = 0;
if (!vm::check_addr(addr))
if (virtual_load)
{
addr = std::exchange(allocating_address, allocating_address + utils::align<u32>(mem_size, 0x10000));
}
else
{
addr = (!ar ? vm::alloc(mem_size, vm::main) : ar->operator u32());
}
_seg.ptr = vm::base(addr);
if (virtual_load)
{
// Leave additional room for the analyser so it can safely access beyond limit a bit
// Because with VM the address sapce is not really a limit so any u32 address is valid there, here it is UB to create pointer that goes beyond the boundaries
// TODO: Use make_shared_for_overwrite when all compilers support it
const usz alloc_size = utils::align<usz>(mem_size, 0x10000) + 4096;
prx->allocations.push_back(std::shared_ptr<u8[]>(new u8[alloc_size]));
_seg.ptr = prx->allocations.back().get();
std::memset(static_cast<u8*>(_seg.ptr) + prog.bin.size(), 0, alloc_size - 4096 - prog.bin.size());
}
else if (!vm::check_addr(addr))
{
fmt::throw_exception("vm::alloc() failed (size=0x%x)", mem_size);
}
_seg.addr = addr;
_seg.size = mem_size;
_seg.filesz = file_size;
prx->addr_to_seg_index.emplace(addr, prx->segs.size() - 1);
// Copy segment data
if (!ar) std::memcpy(vm::base(addr), prog.bin.data(), file_size);
if (!ar) std::memcpy(ensure(prx->get_ptr<void>(addr)), prog.bin.data(), file_size);
ppu_loader.warning("**** Loaded to 0x%x...0x%x (size=0x%x)", addr, addr + mem_size - 1, mem_size);
// Hash segment
@ -1312,7 +1347,7 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
sha1_update(&sha, prog.bin.data(), prog.bin.size());
// Initialize executable code if necessary
if (prog.p_flags & 0x1)
if (prog.p_flags & 0x1 && !virtual_load)
{
if (ar)
{
@ -1322,10 +1357,6 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
ppu_register_range(addr, mem_size);
}
_seg.addr = addr;
_seg.size = mem_size;
_seg.filesz = file_size;
}
break;
@ -1421,63 +1452,63 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
{
case 1: // R_PPC64_ADDR32
{
const u32 value = vm::_ref<u32>(raddr) = static_cast<u32>(rdata);
const u32 value = *ensure(prx->get_ptr<u32>(raddr)) = static_cast<u32>(rdata);
ppu_loader.trace("**** RELOCATION(1): 0x%x <- 0x%08x (0x%llx)", raddr, value, rdata);
break;
}
case 4: //R_PPC64_ADDR16_LO
{
const u16 value = vm::_ref<u16>(raddr) = static_cast<u16>(rdata);
const u16 value = *ensure(prx->get_ptr<u16>(raddr)) = static_cast<u16>(rdata);
ppu_loader.trace("**** RELOCATION(4): 0x%x <- 0x%04x (0x%llx)", raddr, value, rdata);
break;
}
case 5: //R_PPC64_ADDR16_HI
{
const u16 value = vm::_ref<u16>(raddr) = static_cast<u16>(rdata >> 16);
const u16 value = *ensure(prx->get_ptr<u16>(raddr)) = static_cast<u16>(rdata >> 16);
ppu_loader.trace("**** RELOCATION(5): 0x%x <- 0x%04x (0x%llx)", raddr, value, rdata);
break;
}
case 6: //R_PPC64_ADDR16_HA
{
const u16 value = vm::_ref<u16>(raddr) = static_cast<u16>(rdata >> 16) + (rdata & 0x8000 ? 1 : 0);
const u16 value = *ensure(prx->get_ptr<u16>(raddr)) = static_cast<u16>(rdata >> 16) + (rdata & 0x8000 ? 1 : 0);
ppu_loader.trace("**** RELOCATION(6): 0x%x <- 0x%04x (0x%llx)", raddr, value, rdata);
break;
}
case 10: //R_PPC64_REL24
{
const u32 value = vm::_ref<ppu_bf_t<be_t<u32>, 6, 24>>(raddr) = static_cast<u32>(rdata - raddr) >> 2;
const u32 value = *ensure(prx->get_ptr<ppu_bf_t<be_t<u32>, 6, 24>>(raddr)) = static_cast<u32>(rdata - raddr) >> 2;
ppu_loader.warning("**** RELOCATION(10): 0x%x <- 0x%06x (0x%llx)", raddr, value, rdata);
break;
}
case 11: //R_PPC64_REL14
{
const u32 value = vm::_ref<ppu_bf_t<be_t<u32>, 16, 14>>(raddr) = static_cast<u32>(rdata - raddr) >> 2;
const u32 value = *ensure(prx->get_ptr<ppu_bf_t<be_t<u32>, 16, 14>>(raddr)) = static_cast<u32>(rdata - raddr) >> 2;
ppu_loader.warning("**** RELOCATION(11): 0x%x <- 0x%06x (0x%llx)", raddr, value, rdata);
break;
}
case 38: //R_PPC64_ADDR64
{
const u64 value = vm::_ref<u64>(raddr) = rdata;
const u64 value = *ensure(prx->get_ptr<u64>(raddr)) = rdata;
ppu_loader.trace("**** RELOCATION(38): 0x%x <- 0x%016llx (0x%llx)", raddr, value, rdata);
break;
}
case 44: //R_PPC64_REL64
{
const u64 value = vm::_ref<u64>(raddr) = rdata - raddr;
const u64 value = *ensure(prx->get_ptr<u64>(raddr)) = rdata - raddr;
ppu_loader.trace("**** RELOCATION(44): 0x%x <- 0x%016llx (0x%llx)", raddr, value, rdata);
break;
}
case 57: //R_PPC64_ADDR16_LO_DS
{
const u16 value = vm::_ref<ppu_bf_t<be_t<u16>, 0, 14>>(raddr) = static_cast<u16>(rdata) >> 2;
const u16 value = *ensure(prx->get_ptr<ppu_bf_t<be_t<u16>, 0, 14>>(raddr)) = static_cast<u16>(rdata) >> 2;
ppu_loader.trace("**** RELOCATION(57): 0x%x <- 0x%04x (0x%llx)", raddr, value, rdata);
break;
}
@ -1512,7 +1543,7 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
};
// Access library information (TODO)
const vm::cptr<ppu_prx_library_info> lib_info = vm::cast(prx->segs[0].addr + elf.progs[0].p_paddr - elf.progs[0].p_offset);
const auto lib_info = ensure(prx->get_ptr<const ppu_prx_library_info>(prx->segs[0].addr + elf.progs[0].p_paddr - elf.progs[0].p_offset));
const std::string lib_name = lib_info->name;
strcpy_trunc(prx->module_info_name, lib_name);
@ -1523,7 +1554,7 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
prx->exports_start = lib_info->exports_start;
prx->exports_end = lib_info->exports_end;
for (usz start = prx->exports_start, size = 0;; size++, start += vm::read8(start) ? vm::read8(start) : sizeof(ppu_prx_module_info))
for (usz start = prx->exports_start, size = 0;; size++)
{
if (start >= prx->exports_end)
{
@ -1531,18 +1562,25 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
prx->m_external_loaded_flags.resize(size);
break;
}
const u8 increment = *ensure(prx->get_ptr<u8>(start));
start += increment ? increment : sizeof(ppu_prx_module_info);
}
ppu_loader.warning("Library %s (rtoc=0x%x):", lib_name, lib_info->toc);
prx->specials = ppu_load_exports(&link, prx->exports_start, prx->exports_end, true);
prx->imports = ppu_load_imports(prx->relocs, &link, lib_info->imports_start, lib_info->imports_end);
if (!virtual_load)
{
prx->specials = ppu_load_exports(&link, prx->exports_start, prx->exports_end, true);
prx->imports = ppu_load_imports(prx->relocs, &link, lib_info->imports_start, lib_info->imports_end);
}
std::stable_sort(prx->relocs.begin(), prx->relocs.end());
toc = lib_info->toc;
}
else
{
ppu_loader.fatal("Library %s: PRX library info not found");
ppu_loader.error("Library %s: PRX library info not found");
}
prx->start.set(prx->specials[0xbc9a0086]);
@ -1579,12 +1617,12 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
const std::string hash_seg = fmt::format("%s-%u", hash, i);
// Apply the patch
auto _applied = g_fxo->get<patch_engine>().apply(hash_seg, vm::get_super_ptr(seg.addr), seg.size);
auto _applied = g_fxo->get<patch_engine>().apply(hash_seg, [&](u32 addr) { return prx->get_ptr<u8>(addr + seg.addr); }, seg.size);
if (!Emu.GetTitleID().empty())
{
// Alternative patch
_applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash_seg, vm::get_super_ptr(seg.addr), seg.size);
_applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash_seg, [&](u32 addr) { return prx->get_ptr<u8>(addr + seg.addr); }, seg.size);
}
// Rebase patch offsets
@ -1605,12 +1643,15 @@ std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, const std::stri
// Embedded SPU elf patching
for (const auto& seg : prx->segs)
{
ppu_check_patch_spu_images(seg);
ppu_check_patch_spu_images(*prx, seg);
}
prx->analyse(toc, 0, end, applied);
try_spawn_ppu_if_exclusive_program(*prx);
if (!ar && !virtual_load)
{
try_spawn_ppu_if_exclusive_program(*prx);
}
return prx;
}
@ -1667,7 +1708,10 @@ void ppu_unload_prx(const lv2_prx& prx)
{
if (!seg.size) continue;
vm::dealloc(seg.addr, vm::main);
if (seg.ptr == vm::base(seg.addr))
{
vm::dealloc(seg.addr, vm::main);
}
const std::string hash_seg = fmt::format("%s-%u", hash, &seg - prx.segs.data());
@ -1682,7 +1726,7 @@ void ppu_unload_prx(const lv2_prx& prx)
}
}
bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, utils::serial* ar)
bool ppu_load_exec(const ppu_exec_object& elf, bool virtual_load, const std::string& elf_path, utils::serial* ar)
{
if (elf != elf_error::ok)
{
@ -1753,6 +1797,12 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
} error_handler{_main};
if (virtual_load)
{
// No need for cleanup
error_handler.errored = false;
}
// Allocate memory at fixed positions
for (const auto& prog : elf.progs)
{
@ -1774,13 +1824,25 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
{
if (prog.bin.size() > size || prog.bin.size() != prog.p_filesz)
{
ppu_loader.fatal("ppu_load_exec(): Invalid binary size (0x%llx, memsz=0x%x)", prog.bin.size(), size);
ppu_loader.error("ppu_load_exec(): Invalid binary size (0x%llx, memsz=0x%x)", prog.bin.size(), size);
return false;
}
const bool already_loaded = ar && vm::check_addr(addr, vm::page_readable, size);
if (already_loaded)
_seg.ptr = vm::base(addr);
if (virtual_load)
{
// Leave additional room for the analyser so it can safely access beyond limit a bit
// Because with VM the address sapce is not really a limit so any u32 address is valid there, here it is UB to create pointer that goes beyond the boundaries
// TODO: Use make_shared_for_overwrite when all compilers support it
const usz alloc_size = utils::align<usz>(size, 0x10000) + 4096;
_main.allocations.push_back(std::shared_ptr<u8[]>(new u8[alloc_size]));
_seg.ptr = _main.allocations.back().get();
std::memset(static_cast<u8*>(_seg.ptr) + prog.bin.size(), 0, alloc_size - 4096 - prog.bin.size());
}
else if (already_loaded)
{
}
else if (!vm::falloc(addr, size, vm::main))
@ -1789,15 +1851,19 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
if (!vm::falloc(addr, size))
{
ppu_loader.fatal("ppu_load_exec(): vm::falloc() failed (addr=0x%x, memsz=0x%x)", addr, size);
ppu_loader.error("ppu_load_exec(): vm::falloc() failed (addr=0x%x, memsz=0x%x)", addr, size);
return false;
}
}
// Store only LOAD segments (TODO)
_main.segs.emplace_back(_seg);
_main.addr_to_seg_index.emplace(addr, _main.segs.size() - 1);
// Copy segment data, hash it
if (!already_loaded)
{
std::memcpy(vm::base(addr), prog.bin.data(), prog.bin.size());
std::memcpy(_main.get_ptr<void>(addr), prog.bin.data(), prog.bin.size());
}
else
{
@ -1812,7 +1878,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
sha1_update(&sha, prog.bin.data(), prog.bin.size());
// Initialize executable code if necessary
if (prog.p_flags & 0x1)
if (prog.p_flags & 0x1 && !virtual_load)
{
if (already_loaded && ar)
{
@ -1822,9 +1888,6 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
ppu_register_range(addr, size);
}
// Store only LOAD segments (TODO)
_main.segs.emplace_back(_seg);
}
}
@ -1868,12 +1931,12 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
Emu.SetExecutableHash(hash);
// Apply the patch
auto applied = g_fxo->get<patch_engine>().apply(!ar ? hash : std::string{}, vm::g_base_addr);
auto applied = g_fxo->get<patch_engine>().apply(!ar ? hash : std::string{}, [&](u32 addr) { return _main.get_ptr<u8>(addr); });
if (!ar && !Emu.GetTitleID().empty())
{
// Alternative patch
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, vm::g_base_addr);
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, [&](u32 addr) { return _main.get_ptr<u8>(addr); });
}
if (applied.empty())
@ -1891,11 +1954,11 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
// Embedded SPU elf patching
for (const auto& seg : _main.segs)
{
ppu_check_patch_spu_images(seg);
ppu_check_patch_spu_images(_main, seg);
}
// Static HLE patching
if (g_cfg.core.hook_functions)
if (g_cfg.core.hook_functions && !virtual_load)
{
auto shle = g_fxo->init<statichle_handler>(0);
@ -1943,7 +2006,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
if ((prog.p_vaddr | prog.p_filesz | prog.p_memsz) > u32{umax})
{
ppu_loader.fatal("ppu_load_exec(): TLS segment is invalid!");
ppu_loader.error("ppu_load_exec(): TLS segment is invalid!");
return false;
}
@ -1970,7 +2033,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
//be_t<u32> crash_dump_param_addr;
};
const auto& info = vm::_ref<process_param_t>(vm::cast(prog.p_vaddr));
const auto& info = *ensure(_main.get_ptr<process_param_t>(vm::cast(prog.p_vaddr)));
if (info.size < sizeof(process_param_t))
{
@ -2025,7 +2088,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
be_t<u32> unk2;
};
const auto& proc_prx_param = vm::_ref<const ppu_proc_prx_param_t>(vm::cast(prog.p_vaddr));
const auto& proc_prx_param = *ensure(_main.get_ptr<const ppu_proc_prx_param_t>(vm::cast(prog.p_vaddr)));
ppu_loader.notice("* libent_start = *0x%x", proc_prx_param.libent_start);
ppu_loader.notice("* libstub_start = *0x%x", proc_prx_param.libstub_start);
@ -2034,12 +2097,16 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
if (proc_prx_param.magic != 0x1b434cecu)
{
ppu_loader.fatal("ppu_load_exec(): Bad magic! (0x%x)", proc_prx_param.magic);
ppu_loader.error("ppu_load_exec(): Bad magic! (0x%x)", proc_prx_param.magic);
return false;
}
ppu_load_exports(&link, proc_prx_param.libent_start, proc_prx_param.libent_end);
ppu_load_imports(_main.relocs, &link, proc_prx_param.libstub_start, proc_prx_param.libstub_end);
if (!virtual_load)
{
ppu_load_exports(&link, proc_prx_param.libent_start, proc_prx_param.libent_end);
ppu_load_imports(_main.relocs, &link, proc_prx_param.libstub_start, proc_prx_param.libstub_end);
}
std::stable_sort(_main.relocs.begin(), _main.relocs.end());
}
break;
@ -2106,7 +2173,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
load_libs.emplace("libsysmodule.sprx");
}
if (ar || Emu.IsVsh())
if (ar || Emu.IsVsh() || virtual_load)
{
// Cannot be used with vsh.self or savestates (they self-manage itself)
load_libs.clear();
@ -2149,8 +2216,15 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
g_fxo->init<lv2_memory_container>(mem_size);
}
void init_fxo_for_exec(utils::serial* ar, bool full);
init_fxo_for_exec(ar, false);
if (!virtual_load)
{
void init_fxo_for_exec(utils::serial* ar, bool full);
init_fxo_for_exec(ar, false);
}
else
{
Emu.ConfigurePPUCache();
}
liblv2_begin = 0;
liblv2_end = 0;
@ -2165,13 +2239,13 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
{
ppu_loader.warning("Loading library: %s", name);
auto prx = ppu_load_prx(obj, lle_dir + name, 0, nullptr);
auto prx = ppu_load_prx(obj, false, lle_dir + name, 0, nullptr);
prx->state = PRX_STATE_STARTED;
prx->load_exports();
if (prx->funcs.empty())
{
ppu_loader.fatal("Module %s has no functions!", name);
ppu_loader.error("Module %s has no functions!", name);
}
else
{
@ -2196,7 +2270,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
}
}
if (ar)
if (ar || virtual_load)
{
error_handler.errored = false;
return true;
@ -2332,7 +2406,7 @@ bool ppu_load_exec(const ppu_exec_object& elf, const std::string& elf_path, util
return true;
}
std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object& elf, const std::string& path, s64 file_offset, utils::serial* ar)
std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object& elf, bool virtual_load, const std::string& path, s64 file_offset, utils::serial* ar)
{
if (elf != elf_error::ok)
{
@ -2396,11 +2470,23 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
const bool already_loaded = !!ar; // Unimplemented optimization for savestates
if (already_loaded)
_seg.ptr = vm::base(addr);
if (virtual_load)
{
// Leave additional room for the analyser so it can safely access beyond limit a bit
// Because with VM the address sapce is not really a limit so any u32 address is valid there, here it is UB to create pointer that goes beyond the boundaries
// TODO: Use make_shared_for_overwrite when all compilers support it
const usz alloc_size = utils::align<usz>(size, 0x10000) + 4096;
ovlm->allocations.push_back(std::shared_ptr<u8[]>(new u8[alloc_size]));
_seg.ptr = ovlm->allocations.back().get();
std::memset(static_cast<u8*>(_seg.ptr) + prog.bin.size(), 0, alloc_size - 4096 - prog.bin.size());
}
else if (already_loaded)
{
if (!vm::check_addr(addr, vm::page_readable, size))
{
ppu_loader.fatal("ppu_load_overlay(): Archived PPU overlay memory has not been found! (addr=0x%x, memsz=0x%x)", addr, size);
ppu_loader.error("ppu_load_overlay(): Archived PPU overlay memory has not been found! (addr=0x%x, memsz=0x%x)", addr, size);
return {nullptr, CELL_EABORT};
}
}
@ -2418,14 +2504,18 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
return {nullptr, CELL_EBUSY};
}
// Store only LOAD segments (TODO)
ovlm->segs.emplace_back(_seg);
ovlm->addr_to_seg_index.emplace(addr, ovlm->segs.size() - 1);
// Copy segment data, hash it
if (!already_loaded) std::memcpy(vm::base(addr), prog.bin.data(), prog.bin.size());
if (!already_loaded) std::memcpy(ensure(ovlm->get_ptr<void>(addr)), prog.bin.data(), prog.bin.size());
sha1_update(&sha, reinterpret_cast<const uchar*>(&prog.p_vaddr), sizeof(prog.p_vaddr));
sha1_update(&sha, reinterpret_cast<const uchar*>(&prog.p_memsz), sizeof(prog.p_memsz));
sha1_update(&sha, prog.bin.data(), prog.bin.size());
// Initialize executable code if necessary
if (prog.p_flags & 0x1)
if (prog.p_flags & 0x1 && !virtual_load)
{
if (ar)
{
@ -2435,9 +2525,6 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
ppu_register_range(addr, size);
}
// Store only LOAD segments (TODO)
ovlm->segs.emplace_back(_seg);
}
}
@ -2479,18 +2566,18 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
}
// Apply the patch
auto applied = g_fxo->get<patch_engine>().apply(hash, vm::g_base_addr);
auto applied = g_fxo->get<patch_engine>().apply(hash, [ovlm](u32 addr) { return ovlm->get_ptr<u8>(addr); });
if (!Emu.GetTitleID().empty())
{
// Alternative patch
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, vm::g_base_addr);
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, [ovlm](u32 addr) { return ovlm->get_ptr<u8>(addr); });
}
// Embedded SPU elf patching
for (const auto& seg : ovlm->segs)
{
ppu_check_patch_spu_images(seg);
ppu_check_patch_spu_images(*ovlm, seg);
}
if (applied.empty())
@ -2523,7 +2610,7 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
//and a lot of zeros.
};
const auto& info = vm::_ref<process_param_t>(vm::cast(prog.p_vaddr));
const auto& info = *ensure(ovlm->get_ptr<process_param_t>(vm::cast(prog.p_vaddr)));
if (info.size < sizeof(process_param_t))
{
@ -2561,7 +2648,7 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
be_t<u32> unk2;
};
const auto& proc_prx_param = vm::_ref<const ppu_proc_prx_param_t>(vm::cast(prog.p_vaddr));
const auto& proc_prx_param = *ensure(ovlm->get_ptr<const ppu_proc_prx_param_t>(vm::cast(prog.p_vaddr)));
ppu_loader.notice("* libent_start = *0x%x", proc_prx_param.libent_start);
ppu_loader.notice("* libstub_start = *0x%x", proc_prx_param.libstub_start);
@ -2573,8 +2660,11 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
fmt::throw_exception("Bad magic! (0x%x)", proc_prx_param.magic);
}
ppu_load_exports(&link, proc_prx_param.libent_start, proc_prx_param.libent_end);
ppu_load_imports(ovlm->relocs, &link, proc_prx_param.libstub_start, proc_prx_param.libstub_end);
if (!virtual_load)
{
ppu_load_exports(&link, proc_prx_param.libent_start, proc_prx_param.libent_end);
ppu_load_imports(ovlm->relocs, &link, proc_prx_param.libstub_start, proc_prx_param.libstub_end);
}
}
break;
}
@ -2593,7 +2683,7 @@ std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_ex
// Validate analyser results (not required)
ovlm->validate(0);
if (!ar)
if (!ar && !virtual_load)
{
idm::import_existing<lv2_obj, lv2_overlay>(ovlm);
try_spawn_ppu_if_exclusive_program(*ovlm);
@ -2641,7 +2731,7 @@ bool ppu_load_rel_exec(const ppu_rel_object& elf)
if (!addr)
{
ppu_loader.fatal("ppu_load_rel_exec(): vm::alloc() failed (memsz=0x%x)", memsize);
ppu_loader.error("ppu_load_rel_exec(): vm::alloc() failed (memsz=0x%x)", memsize);
return false;
}

151
rpcs3/Emu/Cell/PPUThread.cpp
View File

@ -161,9 +161,10 @@ extern void ppu_initialize();
extern void ppu_finalize(const ppu_module& info);
extern bool ppu_initialize(const ppu_module& info, bool = false);
static void ppu_initialize2(class jit_compiler& jit, const ppu_module& module_part, const std::string& cache_path, const std::string& obj_name);
extern std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object&, const std::string& path, s64 file_offset, utils::serial* = nullptr);
extern bool ppu_load_exec(const ppu_exec_object&, bool virtual_load, const std::string&, utils::serial* = nullptr);
extern std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object&, bool virtual_load, const std::string& path, s64 file_offset, utils::serial* = nullptr);
extern void ppu_unload_prx(const lv2_prx&);
extern std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object&, const std::string&, s64 file_offset, utils::serial* = nullptr);
extern std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object&, bool virtual_load, const std::string&, s64 file_offset, utils::serial* = nullptr);
extern void ppu_execute_syscall(ppu_thread& ppu, u64 code);
static void ppu_break(ppu_thread&, ppu_opcode_t, be_t<u32>*, ppu_intrp_func*);
@ -3140,7 +3141,8 @@ extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_
atomic_t<usz> fnext = 0;
shared_mutex sprx_mtx, ovl_mtx;
lf_queue<std::string> possible_exec_file_paths;
shared_mutex ovl_mtx;
named_thread_group workers("SPRX Worker ", std::min<u32>(utils::get_thread_count(), ::size32(file_queue)), [&]
{
@ -3192,17 +3194,11 @@ extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_
if (ppu_prx_object obj = src; (prx_err = obj, obj == elf_error::ok))
{
std::unique_lock lock(sprx_mtx);
if (auto prx = ppu_load_prx(obj, path, offset))
if (auto prx = ppu_load_prx(obj, true, path, offset))
{
lock.unlock();
obj.clear(), src.close(); // Clear decrypted file and elf object memory
ppu_initialize(*prx);
idm::remove<lv2_obj, lv2_prx>(idm::last_id());
lock.lock();
ppu_unload_prx(*prx);
lock.unlock();
ppu_finalize(*prx);
continue;
}
@ -3215,10 +3211,7 @@ extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_
{
while (ovl_err == elf_error::ok)
{
// Only one thread compiles OVL atm, other can compile PRX cuncurrently
std::unique_lock lock(ovl_mtx);
auto [ovlm, error] = ppu_load_overlay(obj, path, offset);
auto [ovlm, error] = ppu_load_overlay(obj, true, path, offset);
if (error)
{
@ -3229,15 +3222,13 @@ extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_
obj.clear(), src.close(); // Clear decrypted file and elf object memory
ppu_initialize(*ovlm);
for (auto& seg : ovlm->segs)
{
vm::dealloc(seg.addr);
// Does not really require this lock, this is done for performance reasons.
// Seems like too many created threads is hard for Windows to manage efficiently with many CPU threads.
std::lock_guard lock(ovl_mtx);
ppu_initialize(*ovlm);
}
lock.unlock();
idm::remove<lv2_obj, lv2_overlay>(idm::last_id());
ppu_finalize(*ovlm);
break;
}
@ -3248,7 +3239,8 @@ extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_
}
}
ppu_log.notice("Failed to precompile '%s' (prx: %s, ovl: %s)", path, prx_err, ovl_err);
ppu_log.notice("Failed to precompile '%s' (prx: %s, ovl: %s): Attempting tratment as executable file", path, prx_err, ovl_err);
possible_exec_file_paths.push(path);
continue;
}
});
@ -3256,6 +3248,97 @@ extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_
// Join every thread
workers.join();
named_thread exec_worker("PPU Exec Worker", [&]
{
if (!possible_exec_file_paths)
{
return;
}
#ifdef __APPLE__
pthread_jit_write_protect_np(false);
#endif
// Set low priority
thread_ctrl::scoped_priority low_prio(-1);
auto slice = possible_exec_file_paths.pop_all();
auto main_module = std::move(g_fxo->get<main_ppu_module>());
for (; slice; slice.pop_front(), g_progr_fdone++)
{
g_progr_ftotal++;
if (Emu.IsStopped())
{
continue;
}
const std::string path = *slice;
ppu_log.notice("Trying to load as executable: %s", path);
// Load MSELF, SPRX or SELF
fs::file src{path};
if (!src)
{
ppu_log.error("Failed to open '%s' (%s)", path, fs::g_tls_error);
continue;
}
// Some files may fail to decrypt due to the lack of klic
src = decrypt_self(std::move(src));
if (!src)
{
ppu_log.notice("Failed to decrypt '%s'", path);
continue;
}
elf_error exec_err{};
if (ppu_exec_object obj = src; (exec_err = obj, obj == elf_error::ok))
{
while (exec_err == elf_error::ok)
{
if (!ppu_load_exec(obj, true, path))
{
// Abort
exec_err = elf_error::header_type;
break;
}
main_ppu_module& _main = g_fxo->get<main_ppu_module>();
if (!_main.analyse(0, _main.elf_entry, _main.seg0_code_end, _main.applied_pathes, [](){ return Emu.IsStopped(); }))
{
break;
}
obj.clear(), src.close(); // Clear decrypted file and elf object memory
ppu_initialize(_main);
ppu_finalize(_main);
_main = {};
break;
}
if (exec_err == elf_error::ok)
{
continue;
}
}
ppu_log.notice("Failed to precompile '%s' as executable (%s)", path, exec_err);
continue;
}
g_fxo->get<main_ppu_module>() = std::move(main_module);
});
exec_worker();
// Revert changes
if (!had_ovl)
@ -3360,7 +3443,7 @@ extern void ppu_initialize()
{
const std::string eseibrd = mount_point + "/vsh/module/eseibrd.sprx";
if (auto prx = ppu_load_prx(ppu_prx_object{decrypt_self(fs::file{eseibrd})}, eseibrd, 0))
if (auto prx = ppu_load_prx(ppu_prx_object{decrypt_self(fs::file{eseibrd})}, true, eseibrd, 0))
{
// Check if cache exists for this infinitesimally small prx
dev_flash_located = ppu_initialize(*prx, true);
@ -3563,9 +3646,11 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
continue;
}
for (u32 i = addr; i < addr + size; i += 4)
auto i_ptr = ensure(info.get_ptr<u32>(addr));
for (u32 i = addr; i < addr + size; i += 4, i_ptr++)
{
if (g_ppu_itype.decode(vm::read32(i)) == ppu_itype::MFVSCR)
if (g_ppu_itype.decode(*i_ptr) == ppu_itype::MFVSCR)
{
ppu_log.warning("MFVSCR found");
has_mfvscr = true;
@ -3708,7 +3793,7 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
if (roff > addr)
{
// Hash from addr to the beginning of the relocation
sha1_update(&ctx, vm::_ptr<const u8>(addr), roff - addr);
sha1_update(&ctx, ensure(info.get_ptr<const u8>(addr)), roff - addr);
}
// Hash relocation type instead
@ -3721,9 +3806,11 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
if (has_dcbz == 1)
{
for (u32 i = addr, end = block.second + block.first - 1; i <= end; i += 4)
auto i_ptr = ensure(info.get_ptr<u32>(addr));
for (u32 i = addr, end = block.second + block.first - 1; i <= end; i += 4, i_ptr++)
{
if (g_ppu_itype.decode(vm::read32(i)) == ppu_itype::DCBZ)
if (g_ppu_itype.decode(*i_ptr) == ppu_itype::DCBZ)
{
has_dcbz = 2;
break;
@ -3732,7 +3819,7 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
}
// Hash from addr to the end of the block
sha1_update(&ctx, vm::_ptr<const u8>(addr), block.second - (addr - block.first));
sha1_update(&ctx, ensure(info.get_ptr<const u8>(addr)), block.second - (addr - block.first));
}
if (reloc)
@ -3742,9 +3829,11 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
if (has_dcbz == 1)
{
for (u32 i = func.addr, end = func.addr + func.size - 1; i <= end; i += 4)
auto i_ptr = ensure(info.get_ptr<u32>(func.addr));
for (u32 i = func.addr, end = func.addr + func.size - 1; i <= end; i += 4, i_ptr++)
{
if (g_ppu_itype.decode(vm::read32(i)) == ppu_itype::DCBZ)
if (g_ppu_itype.decode(*i_ptr) == ppu_itype::DCBZ)
{
has_dcbz = 2;
break;
@ -3752,7 +3841,7 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
}
}
sha1_update(&ctx, vm::_ptr<const u8>(func.addr), func.size);
sha1_update(&ctx, ensure(info.get_ptr<const u8>(func.addr)), func.size);
}
if (false)

6
rpcs3/Emu/Cell/PPUTranslator.cpp
View File

@ -150,7 +150,7 @@ Function* PPUTranslator::Translate(const ppu_function& info)
for (u32 addr = m_addr; addr < m_addr + info.size; addr += 4)
{
const u32 op = vm::read32(vm::cast(addr + base));
const u32 op = *ensure(m_info.get_ptr<u32>(addr + base));
switch (g_ppu_itype.decode(op))
{
@ -214,7 +214,7 @@ Function* PPUTranslator::Translate(const ppu_function& info)
const auto block = std::make_pair(info.addr, info.size);
{
// Optimize BLR (prefetch LR)
if (vm::read32(vm::cast(block.first + block.second - 4)) == ppu_instructions::BLR())
if (*ensure(m_info.get_ptr<u32>(block.first + block.second - 4)) == ppu_instructions::BLR())
{
RegLoad(m_lr);
}
@ -239,7 +239,7 @@ Function* PPUTranslator::Translate(const ppu_function& info)
m_rel = nullptr;
}
const u32 op = vm::read32(vm::cast(m_addr + base));
const u32 op = *ensure(m_info.get_ptr<u32>(m_addr + base));
(this->*(s_ppu_decoder.decode(op)))({op});

6
rpcs3/Emu/Cell/lv2/sys_overlay.cpp
View File

@ -13,7 +13,7 @@
#include "sys_overlay.h"
#include "sys_fs.h"
extern std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object&, const std::string& path, s64 file_offset, utils::serial* ar = nullptr);
extern std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object&, bool virtual_load, const std::string& path, s64 file_offset, utils::serial* ar = nullptr);
extern bool ppu_initialize(const ppu_module&, bool = false);
extern void ppu_finalize(const ppu_module&);
@ -43,7 +43,7 @@ static error_code overlay_load_module(vm::ptr<u32> ovlmid, const std::string& vp
return {CELL_ENOEXEC, obj.operator elf_error()};
}
const auto [ovlm, error] = ppu_load_overlay(obj, vfs::get(vpath), file_offset);
const auto [ovlm, error] = ppu_load_overlay(obj, false, vfs::get(vpath), file_offset);
obj.clear();
@ -77,7 +77,7 @@ std::shared_ptr<void> lv2_overlay::load(utils::serial& ar)
{
u128 klic = g_fxo->get<loaded_npdrm_keys>().last_key();
file = make_file_view(std::move(file), offset);
ovlm = ppu_load_overlay(ppu_exec_object{ decrypt_self(std::move(file), reinterpret_cast<u8*>(&klic)) }, path, 0, &ar).first;
ovlm = ppu_load_overlay(ppu_exec_object{ decrypt_self(std::move(file), reinterpret_cast<u8*>(&klic)) }, false, path, 0, &ar).first;
ensure(ovlm);
}
else

6
rpcs3/Emu/Cell/lv2/sys_prx.cpp
View File

@ -17,7 +17,7 @@
#include "sys_memory.h"
#include <span>
extern std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object&, const std::string&, s64, utils::serial* = nullptr);
extern std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object&, bool virtual_load, const std::string&, s64, utils::serial* = nullptr);
extern void ppu_unload_prx(const lv2_prx& prx);
extern bool ppu_initialize(const ppu_module&, bool = false);
extern void ppu_finalize(const ppu_module&);
@ -270,7 +270,7 @@ static error_code prx_load_module(const std::string& vpath, u64 flags, vm::ptr<s
return CELL_PRX_ERROR_ILLEGAL_LIBRARY;
}
const auto prx = ppu_load_prx(obj, path, file_offset);
const auto prx = ppu_load_prx(obj, false, path, file_offset);
obj.clear();
@ -324,7 +324,7 @@ std::shared_ptr<void> lv2_prx::load(utils::serial& ar)
{
u128 klic = g_fxo->get<loaded_npdrm_keys>().last_key();
file = make_file_view(std::move(file), offset);
prx = ppu_load_prx(ppu_prx_object{ decrypt_self(std::move(file), reinterpret_cast<u8*>(&klic)) }, path, 0, &ar);
prx = ppu_load_prx(ppu_prx_object{ decrypt_self(std::move(file), reinterpret_cast<u8*>(&klic)) }, false, path, 0, &ar);
prx->m_loaded_flags = std::move(loaded_flags);
prx->m_external_loaded_flags = std::move(external_flags);

4
rpcs3/Emu/Cell/lv2/sys_spu.cpp
View File

@ -185,12 +185,12 @@ void sys_spu_image::deploy(u8* loc, std::span<const sys_spu_segment> segs, bool
}
// Apply the patch
auto applied = g_fxo->get<patch_engine>().apply(hash, loc);
auto applied = g_fxo->get<patch_engine>().apply(hash, [loc](u32 addr) { return loc + addr; });
if (!Emu.GetTitleID().empty())
{
// Alternative patch
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, loc);
applied += g_fxo->get<patch_engine>().apply(Emu.GetTitleID() + '-' + hash, [loc](u32 addr) { return loc + addr; });
}
(is_verbose ? spu_log.notice : sys_spu.trace)("Loaded SPU image: %s (<- %u)%s", hash, applied.size(), dump);

32
rpcs3/Emu/System.cpp
View File

@ -70,15 +70,15 @@ std::string g_cfg_defaults;
atomic_t<u64> g_watchdog_hold_ctr{0};
extern bool ppu_load_exec(const ppu_exec_object&, const std::string&, utils::serial* = nullptr);
extern bool ppu_load_exec(const ppu_exec_object&, bool virtual_load, const std::string&, utils::serial* = nullptr);
extern void spu_load_exec(const spu_exec_object&);
extern void spu_load_rel_exec(const spu_rel_object&);
extern void ppu_precompile(std::vector<std::string>& dir_queue, std::vector<ppu_module*>* loaded_prx);
extern bool ppu_initialize(const ppu_module&, bool = false);
extern void ppu_finalize(const ppu_module&);
extern void ppu_unload_prx(const lv2_prx&);
extern std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object&, const std::string&, s64 = 0, utils::serial* = nullptr);
extern std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object&, const std::string& path, s64 = 0, utils::serial* = nullptr);
extern std::shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object&, bool virtual_load, const std::string&, s64 = 0, utils::serial* = nullptr);
extern std::pair<std::shared_ptr<lv2_overlay>, CellError> ppu_load_overlay(const ppu_exec_object&, bool virtual_load, const std::string& path, s64 = 0, utils::serial* = nullptr);
extern bool ppu_load_rel_exec(const ppu_rel_object&);
extern bool is_savestate_version_compatible(const std::vector<std::pair<u16, u16>>& data, bool is_boot_check);
extern std::vector<std::pair<u16, u16>> read_used_savestate_versions();
@ -1338,6 +1338,8 @@ game_boot_result Emulator::Load(const std::string& title_id, bool is_disc_patch,
std::vector<std::string> dir_queue;
dir_queue.emplace_back(m_path + '/');
init_fxo_for_exec(nullptr, true);
{
if (m_title_id.empty())
{
@ -1384,7 +1386,7 @@ game_boot_result Emulator::Load(const std::string& title_id, bool is_disc_patch,
const ppu_exec_object obj = src;
if (obj == elf_error::ok && ppu_load_exec(obj, path))
if (obj == elf_error::ok && ppu_load_exec(obj, true, path))
{
g_fxo->get<main_ppu_module>().path = path;
}
@ -1393,22 +1395,6 @@ game_boot_result Emulator::Load(const std::string& title_id, bool is_disc_patch,
sys_log.error("Failed to load binary '%s' (%s)", path, obj.get_error());
}
}
else
{
// Workaround for analyser glitches
ensure(vm::falloc(0x10000, 0xf0000, vm::main));
}
}
if (auto& _main = g_fxo->get<main_ppu_module>(); _main.path.empty())
{
init_fxo_for_exec(nullptr, true);
}
if (auto main_ppu = idm::get<named_thread<ppu_thread>>(ppu_thread::id_base))
{
// Created by ppu_load_exec, unwanted
main_ppu->state += cpu_flag::exit;
}
g_fxo->init<named_thread>("SPRX Loader"sv, [this, dir_queue]() mutable
@ -1961,11 +1947,11 @@ game_boot_result Emulator::Load(const std::string& title_id, bool is_disc_patch,
g_fxo->init<main_ppu_module>();
if (ppu_load_exec(ppu_exec, m_path, DeserialManager()))
if (ppu_load_exec(ppu_exec, false, m_path, DeserialManager()))
{
}
// Overlay (OVL) executable (only load it)
else if (vm::map(0x3000'0000, 0x1000'0000, 0x200); !ppu_load_overlay(ppu_exec, m_path).first)
else if (vm::map(0x3000'0000, 0x1000'0000, 0x200); !ppu_load_overlay(ppu_exec, false, m_path).first)
{
ppu_exec.set_error(elf_error::header_type);
}
@ -1989,7 +1975,7 @@ game_boot_result Emulator::Load(const std::string& title_id, bool is_disc_patch,
// PPU PRX
GetCallbacks().on_ready();
g_fxo->init(false);
ppu_load_prx(ppu_prx, m_path);
ppu_load_prx(ppu_prx, false, m_path);
Pause(true);
}
else if (spu_exec.open(elf_file) == elf_error::ok)