raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

graphics: bound unbounded buffer reads and writes (#1066)

Browse Source 
Co-authored-by: João Lobo <joaodlobo@gmail.com>
This commit is contained in:
nmzik
2026-10-05 23:08:28 +04:00
parent 984010227f
commit e5dcfdb87b
4 changed files with 123 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/pipeline/descriptors.cpp
+44 -22
View File
@@ -807,30 +807,52 @@ void RenderExecutor::PrepareBindings(const ShaderStageRuntime& runtime,
}
}
void RenderExecutor::FindBuffers(PreparedBindings& prepared) {
void RenderExecutor::FindBuffers(std::span<PreparedBindings* const> stages) {
KYTY_PROFILER_FUNCTION();
EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
const auto& program = *prepared.runtime->program;
const auto& snapshot = *prepared.runtime->resources;
auto& cache = m_context.GetBufferCache();
prepared.buffer_sources.clear();
const auto& layout = program.bindings;
if (layout.memory_offset_count == 0) {
return;
}
const auto& resources = layout.descriptors.front().resources;
prepared.buffer_sources.reserve(resources.size());
for (const auto resource: resources) {
auto descriptor = DecodeNativeDescriptor<ShaderBufferResource>(snapshot.buffers[resource]);
const auto address = descriptor.Base48();
const auto requested_size = descriptor.GetSize();
if (address == 0 || requested_size == 0) {
prepared.buffer_sources.push_back({});
auto& cache = m_context.GetBufferCache();
for (auto* stage: stages) {
auto& prepared = *stage;
EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
const auto& program = *prepared.runtime->program;
const auto& snapshot = *prepared.runtime->resources;
prepared.buffer_sources.clear();
const auto& layout = program.bindings;
if (layout.memory_offset_count == 0) {
continue;
}
const auto size = Libs::LibKernel::Memory::ClampRangeSize(address, requested_size);
prepared.buffer_sources.push_back({address, size, cache.FindBuffer(address, size)});
const auto& resources = layout.descriptors.front().resources;
prepared.buffer_sources.reserve(resources.size());
for (const auto resource: resources) {
const auto descriptor = DecodeNativeDescriptor<ShaderBufferResource>(snapshot.buffers[resource]);
const auto address = descriptor.Base48();
auto size = descriptor.GetSize();
if (address == 0 || size == 0) {
prepared.buffer_sources.push_back({});
continue;
}
if (descriptor.NumRecords() == UINT32_MAX) {
if (program.info.buffers[resource].written) {
// Sentinel write ranges end before tables captured by any bound stage.
for (const auto* reader: stages) {
for (const auto [read_address, read_size]: reader->runtime->resources->specialization_reads) {
if (read_size == 0) continue;
if (read_address > address) {
size = std::min(size, read_address - address);
} else if (address - read_address < read_size) {
EXIT("scalar resource reads overlap a shader buffer write\n");
}
}
}
} else {
const auto& graphics = m_context.GetGraphics();
const auto limit = uint64_t {graphics.GetPhysicalDeviceProperties().limits.maxStorageBufferRange} -
(graphics.StorageMinAlignment() - 1);
size = std::min({size, uint64_t {256} * 1024 * 1024, limit});
}
}
size = Libs::LibKernel::Memory::ClampRangeSize(address, size);
prepared.buffer_sources.push_back({address, size, cache.FindBuffer(address, size)});
}
}
}
@@ -919,8 +941,8 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
void RenderExecutor::PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
std::span<RenderColorInfo> colors) {
bool uses_dma = false;
FindBuffers(stages);
for (auto* stage: stages) {
FindBuffers(*stage);
uses_dma |= stage->runtime->program->info.uses_dma;
}
if (uses_dma) {
src/graphics/host_gpu/renderer/render.h
+1 -1
View File
@@ -161,7 +161,7 @@ public:
uint32_t mode);
void PrepareBindings(const ShaderStageRuntime& runtime, PreparedBindings& prepared);
void FindBuffers(PreparedBindings& bindings);
void FindBuffers(std::span<PreparedBindings* const> stages);
void RebindBuffers(PreparedBindings& bindings);
void RebindImages(PreparedBindings& bindings);
void CommitBindings(CommandBuffer& buffer, vk::PipelineBindPoint pipeline_bind_point,
src/graphics/host_gpu/renderer/renderCompute.cpp
+4 -4
View File
@@ -342,7 +342,8 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
std::copy(std::begin(input_info.dispatch_threads_num), std::end(input_info.dispatch_threads_num),
bindings.shader_data.begin() + program.bindings.dispatch_thread_dword);
}
FindBuffers(bindings);
PreparedBindings* descriptor_stage = &bindings;
FindBuffers(std::span {&descriptor_stage, 1u});
if (program.info.uses_dma) {
m_context.PrepareBda();
}
@@ -351,7 +352,6 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
RebindBuffers(bindings);
auto vk_buffer = buffer.Handle();
PreparedBindings* descriptor_stage = &bindings;
CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline,
std::span {&descriptor_stage, 1u});
bool has_storage_writes = bindings.shared_memory.buffer != nullptr ||
@@ -457,7 +457,8 @@ void RenderExecutor::DispatchIndirect(uint64_t submit_id, CommandBuffer& buffer,
auto& pipeline = m_context.GetPipelineCache().GetComputePipeline(input_info, compute_program);
auto& bindings = m_compute_bindings;
PrepareBindings(input_info.stage, bindings);
FindBuffers(bindings);
PreparedBindings* descriptor_stage = &bindings;
FindBuffers(std::span {&descriptor_stage, 1u});
const auto& program = *input_info.stage.program;
if (program.info.uses_dma) {
m_context.PrepareBda();
@@ -469,7 +470,6 @@ void RenderExecutor::DispatchIndirect(uint64_t submit_id, CommandBuffer& buffer,
args_addr, sizeof(vk::DispatchIndirectCommand), false);
EXIT_IF(args_buffer == nullptr || (args_offset & 3u) != 0);
RebindBuffers(bindings);
PreparedBindings* descriptor_stage = &bindings;
CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline,
std::span {&descriptor_stage, 1u});
const auto vk_buffer = buffer.Handle();