AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

wsl/libhsakmt: refactor ipc implementation

Telusuri Sumber 
Signed-off-by: Flora Cui <flora.cui@amd.com>
Reviewed-by: Tianci Yin <tianci.yin@amd.com>
Part-of: <http://10.67.69.192/wsl/rocr-runtime/-/merge_requests/85>
This commit is contained in:
Flora Cui
2025-07-01 10:42:35 +08:00
committed by Frank Min
orang tua 77093f8f85
melakukan 237377aa02
4 mengubah file dengan 173 tambahan dan 131 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
memory.cpp
+46 -29
Melihat File
@@ -192,7 +192,7 @@ HSAKMT_STATUS hsaKmtAllocMemoryAlignInternal(HSAuint32 PreferredNode,
MemFlags.ui32.CoarseGrain = 1;
// AllocateNonPaged == AllocateIPC
create_info.flags.imported_sys_memfd = !!(MemFlags.ui32.NonPaged && !MemFlags.ui32.GTTAccess);
create_info.flags.sysmem_ipc_sig_exporter = !!(MemFlags.ui32.NonPaged && !MemFlags.ui32.GTTAccess);
create_info.domain = thunk_proxy::AllocDomain::kSystem;
} else {
@@ -207,6 +207,7 @@ HSAKMT_STATUS hsaKmtAllocMemoryAlignInternal(HSAuint32 PreferredNode,
create_info.mem_flags |= thunk_proxy::kKernarg;
create_info.flags.physical_only = MemFlags.ui32.NoAddress;
create_info.flags.alloc_va = !create_info.flags.physical_only;
create_info.flags.interprocess = MemFlags.ui32.NoAddress;
create_info.flags.interprocess |= MemFlags.ui32.Contiguous;
create_info.flags.physical_contiguous = MemFlags.ui32.Contiguous;
@@ -252,7 +253,7 @@ after_trim:
else
*MemoryAddress = reinterpret_cast<void *>(gpu_mem->GpuAddress());
(*allocation_map_)[*MemoryAddress] = Allocation(
(*allocation_map_)[*MemoryAddress] = Allocation(
gpu_mem->GetGpuMemoryHandle(), *MemoryAddress, (uint64_t)*MemoryAddress,
create_info.size, false, nullptr, SizeInBytes,
MemFlags.ui32.GTTAccess ? 0 : PreferredNode, MemFlags.Value);
@@ -460,23 +461,37 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtRegisterGraphicsHandleToNodesExt(HSAuint64 Graphic
uint32_t *gpu_id_array = NULL;
HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;
pr_debug("number of nodes %lu\n", NumberOfNodes);
if (NumberOfNodes == 0) {
RegisterFlags.ui32.requiresVAddr = 0;
NumberOfNodes = 1;
NodeArray = (HSAuint32*)&(dxg_runtime->default_node);
}
if (is_ipc_sysmemfd(GraphicsResourceHandle)) {
GraphicsResourceInfo->NodeId = dxg_runtime->default_node;
pr_info("skip register sysmemfd. It would be released in next step\n");
return HSAKMT_STATUS_SUCCESS;
}
GraphicsResourceInfo->NodeId = NodeArray[0];
return hsaKmtImportDMABufHandle(GraphicsResourceHandle,
GraphicsResourceInfo,
RegisterFlags);
if (NumberOfNodes == 0) {
RegisterFlags.ui32.requiresVAddr = 0;
NumberOfNodes = 1;
NodeArray = (HSAuint32*)&(dxg_runtime->default_node);
}
pr_debug("number of nodes %lu\n", NumberOfNodes);
wsl::thunk::GpuMemoryHandle mem_handle;
ret = import_dmabuf_fd(GraphicsResourceHandle, NodeArray[0],
RegisterFlags.ui32.requiresVAddr,
false, &mem_handle);
if (ret != HSAKMT_STATUS_SUCCESS) {
pr_err("hsaKmtRegisterGraphicsHandleToNodesExt: import_dmabuf_fd failed, "
"GraphicsResourceHandle: %lu, NodeId: %u\n",
GraphicsResourceHandle, NodeArray[0]);
return ret;
}
wsl::thunk::GpuMemory *gpu_mem = wsl::thunk::GpuMemory::Convert(mem_handle);
GraphicsResourceInfo->NodeId = gpu_mem->GetDevice()->NodeId();
GraphicsResourceInfo->SizeInBytes = gpu_mem->ClientSize();
GraphicsResourceInfo->MemoryAddress = RegisterFlags.ui32.requiresVAddr ?
reinterpret_cast<void *>(gpu_mem->GpuAddress()):
reinterpret_cast<void*>(gpu_mem->HandleApeAddress());
return ret;
}
HSAKMT_STATUS HSAKMTAPI hsaKmtExportDMABufHandle(void *MemoryAddress,
@@ -506,25 +521,27 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtExportDMABufHandle(void *MemoryAddress,
}
HSAKMT_STATUS hsaKmtImportDMABufHandle(int DMABufFd,
HsaGraphicsResourceInfo *GraphicsResourceInfo,
HSA_REGISTER_MEM_FLAGS RegisterFlags) {
HSAKMT_STATUS import_dmabuf_fd(int DMABufFd,
uint32_t NodeId,
bool alloc_va,
bool is_ipc_memfd,
wsl::thunk::GpuMemoryHandle *GpuMemHandle) {
CHECK_DXG_OPEN();
wsl::thunk::WDDMDevice* dev = get_wddmdev(GraphicsResourceInfo->NodeId);
*GpuMemHandle = nullptr;
wsl::thunk::WDDMDevice* dev = get_wddmdev(NodeId);
wsl::thunk::GpuMemory *gpu_mem = nullptr;
wsl::thunk::GpuMemoryCreateInfo create_info{};
create_info.dmabuf_fd = DMABufFd;
create_info.flags.imported_vram_alloc_va = RegisterFlags.ui32.requiresVAddr;
create_info.flags.alloc_va = alloc_va;
if (is_ipc_sysmemfd(DMABufFd)) {
if (is_ipc_memfd) {
struct stat st;
fstat(DMABufFd, &st);
uint64_t sz = st.st_size;
if (4096 <= sz && sz < dev->SystemHeapSize() && (sz & 0xfff) == 0) {
pr_debug("DMABufFd %d is sys mem fd(IPC signal), get size:%ld from it\n", DMABufFd, st.st_size);
create_info.flags.imported_sys_memfd = 1; // set to 1 when backend is system memory
create_info.flags.imported_vram_alloc_va = 0; // set to 1 when backend is vram
create_info.flags.sysmem_ipc_sig_importer = 1; // set to 1 when backend is system memory
create_info.size = st.st_size;
}
}
@@ -532,23 +549,23 @@ HSAKMT_STATUS hsaKmtImportDMABufHandle(int DMABufFd,
auto code = dev->CreateGpuMemory(create_info, &gpu_mem);
if (code == ErrorCode::Success) {
void *MemoryAddress;
if (create_info.flags.imported_sys_memfd || create_info.flags.imported_vram_alloc_va)
if (alloc_va)
MemoryAddress = reinterpret_cast<void *>(gpu_mem->GpuAddress());
else
MemoryAddress = reinterpret_cast<void*>(gpu_mem->HandleApeAddress());
*GpuMemHandle = gpu_mem->GetGpuMemoryHandle();
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
/*
* the gpu_mem->Flags() need convert back from GpuMemoryCreateFlags to
* HsaMemFlags, reference hsaKmtAllocMemoryAlign
* */
(*allocation_map_)[MemoryAddress] = Allocation(
gpu_mem->GetGpuMemoryHandle(), MemoryAddress, (uint64_t)MemoryAddress,
*GpuMemHandle, MemoryAddress, (uint64_t)MemoryAddress,
gpu_mem->Size(), false, nullptr, gpu_mem->ClientSize(),
GraphicsResourceInfo->NodeId, gpu_mem->Flags());
NodeId, gpu_mem->Flags());
GraphicsResourceInfo->MemoryAddress = MemoryAddress;
GraphicsResourceInfo->SizeInBytes = gpu_mem->ClientSize();
return HSAKMT_STATUS_SUCCESS;
}
@@ -624,7 +641,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtDeregisterMemory(void *MemoryAddress) {
if (it_ipc != allocation_map_->end()) {
wsl::thunk::GpuMemoryDescFlags flags;
flags.reserved = it_ipc->second.mem_flags_value;
if (flags.is_imported_vram_alloc_va || flags.is_imported_sys_memfd) {
if (flags.is_imported_vram_ipc) {
wsl::thunk::GpuMemory *gpu_mem;
gpu_mem = wsl::thunk::GpuMemory::Convert(it_ipc->second.handle);
allocation_map_->erase(it_ipc);
@@ -677,7 +694,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtMapMemoryToGPUNodes(
if (it_ipc != allocation_map_->end()) {
wsl::thunk::GpuMemoryDescFlags flags;
flags.reserved = it_ipc->second.mem_flags_value;
if (flags.is_imported_vram_alloc_va) {
if (flags.is_imported_vram_ipc) {
wsl::thunk::GpuMemory *gpu_mem;
gpu_mem = wsl::thunk::GpuMemory::Convert(it_ipc->second.handle);
@@ -782,7 +799,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtUnmapMemoryToGPU(void *MemoryAddress) {
if (it_ipc != allocation_map_->end()) {
wsl::thunk::GpuMemoryDescFlags flags;
flags.reserved = it_ipc->second.mem_flags_value;
if (flags.is_imported_vram_alloc_va) {
if (flags.is_imported_vram_ipc) {
wsl::thunk::GpuMemory *gpu_mem;
gpu_mem = wsl::thunk::GpuMemory::Convert(it_ipc->second.handle);