AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add flag for external memory allocations

Browse Source 
ROCr internally uses the same allocation_map_ list to track memory
allocations that are both for internal allocations and allocations by
users of ROCr library. In some edge cases, the library user would call
hsa_amd_pointer_info on an invalid pointer, but ROCR would return the
pointer as valid because this pointer belongs to a memory range that
was allocated internally within ROCr. Adding a flag to differentiate
between internal and external allocations.

Change-Id: I98c52bd85f3985d1ba1b0e3101d2254b003412cf
This commit is contained in:
David Yat Sin
2023-01-04 20:26:25 +00:00
parent 27596aef0c
commit 59685f4492
5 changed files with 28 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
runtime/hsa-runtime/core/runtime/runtime.cpp
+9 -7
View File
@@ -283,14 +283,14 @@ hsa_status_t Runtime::IterateAgent(hsa_status_t (*callback)(hsa_agent_t agent,
}
hsa_status_t Runtime::AllocateMemory(const MemoryRegion* region, size_t size,
MemoryRegion::AllocateFlags alloc_flags,
void** address) {
MemoryRegion::AllocateFlags alloc_flags, void** address,
bool user_request) {
size_t size_requested = size; // region->Allocate(...) may align-up size to granularity
hsa_status_t status = region->Allocate(size, alloc_flags, address);
// Track the allocation result so that it could be freed properly.
if (status == HSA_STATUS_SUCCESS) {
ScopedAcquire<KernelSharedMutex> lock(&memory_lock_);
allocation_map_[*address] = AllocationRegion(region, size, size_requested);
allocation_map_[*address] = AllocationRegion(region, size, size_requested, user_request);
}
return status;
@@ -753,7 +753,7 @@ hsa_status_t Runtime::InteropUnmap(void* ptr) {
hsa_status_t Runtime::PtrInfo(const void* ptr, hsa_amd_pointer_info_t* info, void* (*alloc)(size_t),
uint32_t* num_agents_accessible, hsa_agent_t** accessible,
PtrInfoBlockData* block_info) {
PtrInfoBlockData* block_info, bool user_request) {
static_assert(static_cast<int>(HSA_POINTER_UNKNOWN) == static_cast<int>(HSA_EXT_POINTER_TYPE_UNKNOWN),
"Thunk pointer info mismatch");
static_assert(static_cast<int>(HSA_POINTER_ALLOCATED) == static_cast<int>(HSA_EXT_POINTER_TYPE_HSA),
@@ -824,11 +824,13 @@ hsa_status_t Runtime::PtrInfo(const void* ptr, hsa_amd_pointer_info_t* info, voi
auto fragment = allocation_map_.upper_bound(ptr);
if (fragment != allocation_map_.begin()) {
fragment--;
if ((fragment->first <= ptr) &&
(ptr < reinterpret_cast<const uint8_t*>(fragment->first) + fragment->second.size_requested)) {
if ((fragment->first <= ptr) && (user_request && fragment->second.user_request) &&
(ptr <
reinterpret_cast<const uint8_t*>(fragment->first) + fragment->second.size_requested)) {
// agent and host address must match here. Only lock memory is allowed to have differing
// addresses but lock memory has type HSA_EXT_POINTER_TYPE_LOCKED and cannot be
// suballocated.
retInfo.agentBaseAddress = const_cast<void*>(fragment->first);
retInfo.hostBaseAddress = retInfo.agentBaseAddress;
retInfo.sizeInBytes = fragment->second.size_requested;
@@ -923,7 +925,7 @@ hsa_status_t Runtime::IPCCreate(void* ptr, size_t len, hsa_amd_ipc_memory_t* han
PtrInfoBlockData block;
hsa_amd_pointer_info_t info;
info.size = sizeof(info);
if (PtrInfo(ptr, &info, nullptr, nullptr, nullptr, &block) != HSA_STATUS_SUCCESS)
if (PtrInfo(ptr, &info, nullptr, nullptr, nullptr, &block, true) != HSA_STATUS_SUCCESS)
return HSA_STATUS_ERROR_INVALID_ARGUMENT;
// Temporary: Previous versions of HIP will call hsa_amd_ipc_memory_create with the len aligned to