raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 19:07:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

remove readback hacks, allows to read/write guest memory directly

Browse Source
This commit is contained in:
nmzik
2026-07-17 05:16:00 +02:00
parent 013e9a4485
commit e903c93848
27 changed files with 522 additions and 1228 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/sync.cpp
+30 -105
View File
@@ -13,31 +13,29 @@
#include "kernel/pthread.h"
#include "libs/errno.h"
#include <cstring>
#include <limits>
namespace Libs::Graphics {
constexpr int GRAPHICS_EVENT_QUEUED_GRAPHICS_INTERRUPT = 0x00;
constexpr int GRAPHICS_EVENT_EOP = 0x40;
constexpr uint64_t GRAPHICS_REFERENCE_CLOCK_FREQUENCY = 100000000;
constexpr int GRAPHICS_EVENT_QUEUED_GRAPHICS_INTERRUPT = 0x00;
constexpr int GRAPHICS_EVENT_EOP = 0x40;
constexpr uint64_t GRAPHICS_REFERENCE_CLOCK_FREQUENCY = 100000000;
bool GraphicsScaleReferenceClock(uint64_t host_ticks, uint64_t host_frequency, uint64_t* value) {
if (host_frequency == 0 || value == nullptr) {
return false;
}
const auto whole_seconds = host_ticks / host_frequency;
const auto remainder = host_ticks % host_frequency;
constexpr auto MAX_VALUE = std::numeric_limits<uint64_t>::max();
const auto whole_seconds = host_ticks / host_frequency;
const auto remainder = host_ticks % host_frequency;
constexpr auto MAX_VALUE = std::numeric_limits<uint64_t>::max();
if (whole_seconds > MAX_VALUE / GRAPHICS_REFERENCE_CLOCK_FREQUENCY ||
remainder > MAX_VALUE / GRAPHICS_REFERENCE_CLOCK_FREQUENCY) {
return false;
}
const auto whole_value = whole_seconds * GRAPHICS_REFERENCE_CLOCK_FREQUENCY;
const auto fractional_value =
(remainder * GRAPHICS_REFERENCE_CLOCK_FREQUENCY) / host_frequency;
const auto whole_value = whole_seconds * GRAPHICS_REFERENCE_CLOCK_FREQUENCY;
const auto fractional_value = (remainder * GRAPHICS_REFERENCE_CLOCK_FREQUENCY) / host_frequency;
if (whole_value > MAX_VALUE - fractional_value) {
return false;
}
@@ -50,40 +48,19 @@ uint64_t GraphicsRenderReadReferenceClock() {
const auto host_ticks = LibKernel::KernelReadTsc();
uint64_t value = 0;
if (!GraphicsScaleReferenceClock(host_ticks, host_frequency, &value)) {
EXIT("cannot scale host clock, ticks=0x%016" PRIx64 " frequency=%" PRIu64 "\n",
host_ticks, host_frequency);
EXIT("cannot scale host clock, ticks=0x%016" PRIx64 " frequency=%" PRIu64 "\n", host_ticks,
host_frequency);
}
return value;
}
static void SubmitLabel32(CommandBuffer* buffer, uint32_t* dst, uint32_t value,
LabelCallback callback_1 = nullptr, LabelCallback callback_2 = nullptr,
const uint64_t* args = nullptr) {
auto* label =
LabelCreate32(g_render_ctx->GetGraphicCtx(), dst, value, callback_1, callback_2, args);
static void SubmitLabel(CommandBuffer* buffer, LabelCallback callback_1 = nullptr,
LabelCallback callback_2 = nullptr, const uint64_t* args = nullptr) {
auto* label = LabelCreate(g_render_ctx->GetGraphicCtx(), callback_1, callback_2, args);
LabelSet(buffer, label);
LabelDelete(label);
}
static void SubmitLabel64(CommandBuffer* buffer, uint64_t* dst, uint64_t value,
LabelCallback callback_1 = nullptr, LabelCallback callback_2 = nullptr,
const uint64_t* args = nullptr) {
auto* label =
LabelCreate64(g_render_ctx->GetGraphicCtx(), dst, value, callback_1, callback_2, args);
LabelSet(buffer, label);
LabelDelete(label);
}
template <typename T>
static void PublishImmediateFence(T* dst, T value) {
static_assert(sizeof(T) == sizeof(uint32_t) || sizeof(T) == sizeof(uint64_t));
EXIT_IF(dst == nullptr);
// EOP/EOS fence payloads are constant and published while parsing the packet. Kyty performs the
// direct mapped write under CommandProcessorGuestAccessScope: a protected access is first resolved by
// BufferCache inline, then this instruction retries so the EOP payload wins over prior GPU bytes.
std::memcpy(dst, &value, sizeof(value));
}
static bool CompleteDisplayBufferFlip(const uint64_t* args) {
if (g_render_ctx == nullptr || args == nullptr) {
EXIT("GPU flip completion has invalid state, render_ctx=%p args=%p\n",
@@ -114,8 +91,6 @@ void GraphicsRenderWriteAtEndOfPipe32(uint64_t submit_id, CommandBuffer* buffer,
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWrite), submit_id, 4, value,
0, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
}
void GraphicsRenderWriteAtEndOfPipeGds32(uint64_t submit_id, CommandBuffer* buffer,
@@ -128,28 +103,6 @@ void GraphicsRenderWriteAtEndOfPipeGds32(uint64_t submit_id, CommandBuffer* buff
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWrite), submit_id,
dw_offset, dw_num, 0, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
uint64_t args[LABEL_ARGS_MAX] = {static_cast<uint64_t>(dw_offset),
static_cast<uint64_t>(dw_num),
reinterpret_cast<uint64_t>(dst_gpu_addr), 0};
SubmitLabel32(
buffer, dst_gpu_addr, 0,
[](const uint64_t* args) {
auto dw_offset = static_cast<uint32_t>(args[0]);
auto dw_num = static_cast<uint32_t>(args[1]);
auto* dst_gpu_addr = reinterpret_cast<uint32_t*>(args[2]);
if (dst_gpu_addr == nullptr || dw_num == 0) {
EXIT("invalid asynchronous EOP GDS write\n");
}
const auto size = static_cast<uint64_t>(dw_num) * sizeof(uint32_t);
auto transaction = g_render_ctx->GetGpuResources()->LockAsynchronousGuestWrite(
reinterpret_cast<uint64_t>(dst_gpu_addr), size);
g_render_ctx->GetGdsBuffer()->Read(g_render_ctx->GetGraphicCtx(), dst_gpu_addr,
dw_offset, dw_num);
return false;
},
nullptr, args);
}
void GraphicsRenderWriteAtEndOfPipe64(uint64_t submit_id, CommandBuffer* buffer,
@@ -162,12 +115,10 @@ void GraphicsRenderWriteAtEndOfPipe64(uint64_t submit_id, CommandBuffer* buffer,
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWrite), submit_id, 8,
static_cast<uint32_t>(value), static_cast<uint32_t>(value >> 32u), 0,
reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
}
void GraphicsRenderWriteAtEndOfPipeClockCounter(uint64_t submit_id, CommandBuffer* buffer,
uint64_t* dst_gpu_addr) {
uint64_t* dst_gpu_addr, uint64_t value) {
EXIT_IF(g_render_ctx == nullptr);
EXIT_IF(dst_gpu_addr == nullptr);
EXIT_IF(buffer == nullptr);
@@ -176,8 +127,6 @@ void GraphicsRenderWriteAtEndOfPipeClockCounter(uint64_t submit_id, CommandBuffe
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWrite), submit_id, 8, 0, 0,
0, reinterpret_cast<uint64_t>(dst_gpu_addr));
const auto value = GraphicsRenderReadReferenceClock();
PublishImmediateFence(dst_gpu_addr, value);
LOGF_COLOR(Log::Color::BrightGreen,
"EndOfPipe Signal!!! [0x%016" PRIx64 "] <- Clock: 0x%016" PRIx64 "\n",
reinterpret_cast<uint64_t>(dst_gpu_addr), value);
@@ -185,7 +134,8 @@ void GraphicsRenderWriteAtEndOfPipeClockCounter(uint64_t submit_id, CommandBuffe
void GraphicsRenderWriteAtEndOfPipeClockCounterWithWriteBack(uint64_t submit_id,
CommandBuffer* buffer,
uint64_t* dst_gpu_addr) {
uint64_t* dst_gpu_addr,
uint64_t value) {
EXIT_IF(g_render_ctx == nullptr);
EXIT_IF(dst_gpu_addr == nullptr);
EXIT_IF(buffer == nullptr);
@@ -194,8 +144,6 @@ void GraphicsRenderWriteAtEndOfPipeClockCounterWithWriteBack(uint64_t subm
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWriteBack), submit_id, 8, 0,
0, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
const auto value = GraphicsRenderReadReferenceClock();
PublishImmediateFence(dst_gpu_addr, value);
LOGF_COLOR(Log::Color::BrightGreen,
"EndOfPipe Signal!!! [0x%016" PRIx64 "] <- Clock: 0x%016" PRIx64 "\n",
reinterpret_cast<uint64_t>(dst_gpu_addr), value);
@@ -211,8 +159,6 @@ void GraphicsRenderWriteAtEndOfPipeWithWriteBack64(uint64_t submit_id, CommandBu
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWriteBack), submit_id, 8,
static_cast<uint32_t>(value), static_cast<uint32_t>(value >> 32u), 0,
reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
}
void GraphicsRenderWriteAtEndOfPipeWithWriteBack32(uint64_t submit_id, CommandBuffer* buffer,
@@ -224,8 +170,6 @@ void GraphicsRenderWriteAtEndOfPipeWithWriteBack32(uint64_t submit_id, CommandBu
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWriteBack), submit_id, 4,
value, 0, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
}
void GraphicsRenderWriteAtEndOfPipeWithInterruptWriteBack64(uint64_t submit_id,
@@ -241,12 +185,10 @@ void GraphicsRenderWriteAtEndOfPipeWithInterruptWriteBack64(uint64_t submi
context_id, static_cast<uint32_t>(value),
static_cast<uint32_t>(value >> 32u),
reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
uint64_t args[LABEL_ARGS_MAX] = {context_id != 0 ? context_id : value};
SubmitLabel64(
buffer, nullptr, 0, nullptr,
SubmitLabel(
buffer, nullptr,
[](const uint64_t* args) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->TriggerEopEvent(static_cast<uint32_t>(args[0]));
@@ -266,12 +208,10 @@ void GraphicsRenderWriteAtEndOfPipeWithInterruptWriteBack32(uint64_t submi
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWriteBack), submit_id, 4,
context_id, value, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
uint64_t args[LABEL_ARGS_MAX] = {context_id != 0 ? context_id : value};
SubmitLabel32(
buffer, nullptr, 0, nullptr,
SubmitLabel(
buffer, nullptr,
[](const uint64_t* args) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->TriggerEopEvent(static_cast<uint32_t>(args[0]));
@@ -292,12 +232,10 @@ void GraphicsRenderWriteAtEndOfPipeWithInterrupt64(uint64_t submit_id, CommandBu
context_id, static_cast<uint32_t>(value),
static_cast<uint32_t>(value >> 32u),
reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
uint64_t args[LABEL_ARGS_MAX] = {context_id != 0 ? context_id : value};
SubmitLabel64(
buffer, nullptr, 0, nullptr,
SubmitLabel(
buffer, nullptr,
[](const uint64_t* args) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->TriggerEopEvent(static_cast<uint32_t>(args[0]));
@@ -316,12 +254,10 @@ void GraphicsRenderWriteAtEndOfPipeWithInterrupt32(uint64_t submit_id, CommandBu
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopInterrupt), submit_id, 4,
context_id, value, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
PublishImmediateFence(dst_gpu_addr, value);
uint64_t args[LABEL_ARGS_MAX] = {context_id != 0 ? context_id : value};
SubmitLabel32(
buffer, nullptr, 0, nullptr,
SubmitLabel(
buffer, nullptr,
[](const uint64_t* args) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->TriggerEopEvent(static_cast<uint32_t>(args[0]));
@@ -360,11 +296,9 @@ void GraphicsRenderWriteAtEndOfPipeWithInterruptWriteBackFlip32(
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWriteBackFlip), submit_id,
static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
static_cast<uint32_t>(flip_mode), value, static_cast<uint64_t>(flip_arg));
PublishImmediateFence(dst_gpu_addr, value);
const uint64_t args[LABEL_ARGS_MAX] = {request_id};
SubmitLabel32(
buffer, nullptr, 0, CompleteDisplayBufferFlip,
SubmitLabel(
buffer, CompleteDisplayBufferFlip,
[](const uint64_t* /*args*/) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->TriggerEopEvent(0);
@@ -385,10 +319,8 @@ void GraphicsRenderWriteAtEndOfPipeWithFlip32(uint64_t submit_id, CommandBuffer*
buffer->SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopFlip), submit_id,
static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
static_cast<uint32_t>(flip_mode), value, static_cast<uint64_t>(flip_arg));
PublishImmediateFence(dst_gpu_addr, value);
const uint64_t args[LABEL_ARGS_MAX] = {request_id};
SubmitLabel32(buffer, nullptr, 0, CompleteDisplayBufferFlip, nullptr, args);
SubmitLabel(buffer, CompleteDisplayBufferFlip, nullptr, args);
}
void GraphicsRenderWriteAtEndOfPipeOnlyFlip(uint64_t submit_id, CommandBuffer* buffer, int handle,
@@ -403,7 +335,7 @@ void GraphicsRenderWriteAtEndOfPipeOnlyFlip(uint64_t submit_id, CommandBuffer* b
static_cast<uint32_t>(flip_mode), 0, static_cast<uint64_t>(flip_arg));
const uint64_t args[LABEL_ARGS_MAX] = {request_id};
SubmitLabel32(buffer, nullptr, 0, CompleteDisplayBufferFlip, nullptr, args);
SubmitLabel(buffer, CompleteDisplayBufferFlip, nullptr, args);
}
void GraphicsRenderTriggerEopEventAtEndOfPipe(CommandBuffer* buffer, uint32_t context_id) {
@@ -413,8 +345,8 @@ void GraphicsRenderTriggerEopEventAtEndOfPipe(CommandBuffer* buffer, uint32_t co
uint64_t args[LABEL_ARGS_MAX] = {static_cast<uint64_t>(context_id)};
SubmitLabel32(
buffer, nullptr, 0, nullptr,
SubmitLabel(
buffer, nullptr,
[](const uint64_t* args) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->TriggerEopEvent(static_cast<uint32_t>(args[0]));
@@ -489,13 +421,6 @@ int GraphicsRenderDeleteEqEvent(LibKernel::EventQueue::KernelEqueue eq, int id)
return result;
}
void GraphicsRenderClearGds(uint64_t dw_offset, uint32_t dw_num, uint32_t clear_value) {
EXIT_IF(g_render_ctx == nullptr);
g_render_ctx->GetGdsBuffer()->Clear(g_render_ctx->GetGraphicCtx(), dw_offset, dw_num,
clear_value);
}
void GraphicsRenderReadGds(uint32_t* dst, uint32_t dw_offset, uint32_t dw_size) {
EXIT_IF(g_render_ctx == nullptr);