wsl/libhsakmt: adapt to the new check for kernel object

Signed-off-by: Flora Cui <flora.cui@amd.com>
Reviewed-by: Longlong Yao <Longlong.Yao@amd.com>
Part-of: <http://10.67.69.192/wsl/rocr-runtime/-/merge_requests/99>

memory.cpp

@@ -212,6 +212,9 @@ HSAKMT_STATUS hsaKmtAllocMemoryAlignInternal(HSAuint32 PreferredNode,
   create_info.flags.physical_contiguous = MemFlags.ui32.Contiguous;
   create_info.flags.locked = MemFlags.ui32.NoSubstitute;//AllocatePinned
   create_info.flags.virtual_alloc = MemFlags.ui32.OnlyAddress;
+  create_info.flags.blit_kernel_object =
+      (MemFlags.ui32.ExecuteBlit && MemFlags.ui32.ExecuteAccess &&
+      (create_info.domain == thunk_proxy::AllocDomain::kSystem));
   /*when only alloc virtual or only physical, it's vmm allocation, force to local*/
   if (create_info.flags.virtual_alloc || create_info.flags.physical_only
         || create_info.flags.physical_contiguous) {

wddm/gpu_memory.cpp

@@ -65,6 +65,7 @@ ErrorCode GpuMemory::Init(const GpuMemoryCreateInfo &create_info) {
   desc_.flags.is_imported_sys_memfd = create_info.flags.sysmem_ipc_sig_importer;
   desc_.flags.is_sysmem_exporter = create_info.flags.sysmem_ipc_sig_exporter;
   desc_.flags.is_va_required = create_info.flags.alloc_va;
+  desc_.flags.is_blit_kernel_object = create_info.flags.blit_kernel_object;
 
   /* we can't tell the allocation is regular vmm or ipc mem at creation stage,
      they share same creation parameters, so forcing all vram allocations to

wddm/queue.cpp

@@ -521,15 +521,20 @@ uint32_t ComputeQueue::UpdateIndexStride(uint32_t srd, bool wave32) {
 }
 
 uint64_t ComputeQueue::GetKernelObjAddr(uint64_t addr) const {
-//TODO: convert dev_addr to host_addr
+  /* convert dev_addr to host_addr */
+  auto code = get_gpu_mem((void*)addr);
+  if (code && code->IsBlitKernelObject()) {
+    return code->GpuAddress();
+  }
+
   uint64_t host_addr = 0;
   auto ret = hsakmt_hsa_ven_amd_loader_query_host_address(reinterpret_cast<const void *>(addr),
                                            reinterpret_cast<const void **>(&host_addr));
-  if (ret == HSA_STATUS_ERROR_INVALID_ARGUMENT) {
-    return 0;
+  if (ret == HSA_STATUS_SUCCESS) {
+    return host_addr;
   }
-
-  return host_addr;
+  pr_err("failed to query host address for kernel object %p, ret=%d\n", (void*)addr, ret);
+  return 0;
 }
 
 void ComputeQueue::RingDoorbell() {