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

Correct IPC fragment reuse.

Browse Source 
Memory from the suballocator may be exported via IPC.  If this
happens then the allocating process should not reuse that memory
since it would still be connected to the remote process.  IPC exported
memory must be released back to the driver.

Change-Id: I2ab0c814f63191f753fc3640cc4140ee144bf07f
This commit is contained in:
Sean Keely
2020-04-29 23:12:17 -05:00
parent 09ebc21d13
commit 29b660c91e
5 changed files with 83 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
runtime/hsa-runtime/core/runtime/runtime.cpp
+5
View File
@@ -904,6 +904,11 @@ hsa_status_t Runtime::IPCCreate(void* ptr, size_t len, hsa_amd_ipc_memory_t* han
(reinterpret_cast<uint8_t*>(ptr) - reinterpret_cast<uint8_t*>(block.base)) / 4096;
// Holds size in (4K?) pages in thunk handle: Mark as a fragment and denote offset.
handle->handle[6] |= 0x80000000 | offset;
// Mark block for IPC. Prevents reallocation of exported memory.
ScopedAcquire<KernelMutex> lock(&memory_lock_);
hsa_status_t err = allocation_map_[ptr].region->IPCFragmentExport(ptr);
assert(err == HSA_STATUS_SUCCESS && "Region inconsistent with address map.");
return err;
} else {
if (hsaKmtShareMemory(ptr, len, reinterpret_cast<HsaSharedMemoryHandle*>(handle)) !=
HSAKMT_STATUS_SUCCESS)