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

Properly control lifecycle of ptr info objects

Browse Source 
Buffer mapping to devices and buffer registration to
devices can be changed b/t two pointer info queries.
Thus update buffer mapping info and registration info
only when mapping and registration changed. This is
done by free mapped_node_id_array on mapping to new
device and free registered_node_id_array on registration
and re-allocate them on next ptr info query.

Also uses fmm_mutex to avoid race conditions in case
of calling hsaKmtQueryPointerInfo concurrently with
calling of buffer mapping or registration

Change-Id: Ibc2e20be1fc0147066f873dfa44b21f5015104b7
Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>


[ROCm/ROCR-Runtime commit: f06e887725]
This commit is contained in:
Oak Zeng
2017-11-20 16:15:29 -05:00
parent a62c1f5364
commit 84d88b8e1a
1 changed files with 37 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocr-runtime/src/fmm.c
+37 -7
View File
@@ -1785,6 +1785,13 @@ static int _fmm_map_to_gpu_gtt(manageable_aperture_t *aperture,
object->mapped_device_id_array = temp_mapped_id_array;
object->mapped_device_id_array_size = args.device_ids_array_size;
object->mapping_count = 1;
/* Mapping changed and lifecycle of object->mapped_node_id_array
* terminates here. Free it and allocate on next query
*/
if (object->mapped_node_id_array) {
free(object->mapped_node_id_array);
object->mapped_node_id_array = NULL;
}
exit_ok:
if (!obj)
@@ -1906,6 +1913,14 @@ static int _fmm_map_to_gpu(uint32_t gpu_id, manageable_aperture_t *aperture,
object->mapped_device_id_array = temp_mapped_id_array;
object->mapped_device_id_array_size = args.device_ids_array_size;
/* Mapping changed and lifecycle of object->mapped_node_id_array
* terminates here. Free it and allocate on next query
*/
if (object->mapped_node_id_array) {
free(object->mapped_node_id_array);
object->mapped_node_id_array = NULL;
}
pthread_mutex_unlock(&aperture->fmm_mutex);
if (gpuvm_address) {
@@ -2516,26 +2531,29 @@ HSAKMT_STATUS fmm_register_memory(void *address, uint64_t size_in_bytes,
/* fall through */
}
if (!object) {
pthread_mutex_lock(&aperture->fmm_mutex);
pthread_mutex_lock(&aperture->fmm_mutex);
if (!object)
object = vm_find_object_by_address(aperture, address, 0);
if (!object) {
pthread_mutex_unlock(&aperture->fmm_mutex);
return HSAKMT_STATUS_NOT_SUPPORTED;
}
if (!object)
return HSAKMT_STATUS_NOT_SUPPORTED;
if (object->registered_device_id_array_size > 0) {
/* Multiple registration is allowed, but not changing nodes */
if ((gpu_id_array_size != object->registered_device_id_array_size)
|| memcmp(object->registered_device_id_array,
gpu_id_array, gpu_id_array_size)) {
pr_err("Cannot change nodes in a registered addr.\n");
pthread_mutex_unlock(&aperture->fmm_mutex);
return HSAKMT_STATUS_MEMORY_ALREADY_REGISTERED;
} else {
/* Delete the new array, keep the existing one. */
if (gpu_id_array)
free(gpu_id_array);
pthread_mutex_unlock(&aperture->fmm_mutex);
return HSAKMT_STATUS_SUCCESS;
}
}
@@ -2543,8 +2561,17 @@ HSAKMT_STATUS fmm_register_memory(void *address, uint64_t size_in_bytes,
if (gpu_id_array_size > 0) {
object->registered_device_id_array = gpu_id_array;
object->registered_device_id_array_size = gpu_id_array_size;
/* Registration of object changed. Lifecycle of object->
* registered_node_id_array terminates here. Free old one
* and re-allocate on next query
*/
if (object->registered_node_id_array) {
free(object->registered_node_id_array);
object->registered_node_id_array = NULL;
}
}
pthread_mutex_unlock(&aperture->fmm_mutex);
return HSAKMT_STATUS_SUCCESS;
}
@@ -3024,6 +3051,7 @@ HSAKMT_STATUS fmm_get_mem_info(const void *address, HsaPointerInfo *info)
aperture = fmm_find_aperture(address, NULL);
pthread_mutex_lock(&aperture->fmm_mutex);
vm_obj = vm_find_object_by_address_range(aperture, address);
if (!vm_obj)
vm_obj = vm_find_object_by_userptr_range(aperture, address);
@@ -3051,8 +3079,8 @@ HSAKMT_STATUS fmm_get_mem_info(const void *address, HsaPointerInfo *info)
vm_obj->registered_node_id_array = (uint32_t *)
(uint32_t *)malloc(vm_obj->registered_device_id_array_size);
/* vm_obj->registered_node_id_array allocated here will be
* freed whenever the registration is deregistered or the
* memory being freed
* freed whenever the registration is changed (deregistration or
* register to new nodes) or the memory being freed
*/
for (i = 0; i < info->NRegisteredNodes; i++)
gpuid_to_nodeid(vm_obj->registered_device_id_array[i],
@@ -3066,7 +3094,8 @@ HSAKMT_STATUS fmm_get_mem_info(const void *address, HsaPointerInfo *info)
vm_obj->mapped_node_id_array =
(uint32_t *)malloc(vm_obj->mapped_device_id_array_size);
/* vm_obj->mapped_node_id_array allocated here will be
* freed whenever the mapping is unmapped or memory being freed
* freed whenever the mapping is changed (unmapped or map
* to new nodes) or memory being freed
*/
for (i = 0; i < info->NMappedNodes; i++)
gpuid_to_nodeid(vm_obj->mapped_device_id_array[i],
@@ -3085,6 +3114,7 @@ HSAKMT_STATUS fmm_get_mem_info(const void *address, HsaPointerInfo *info)
}
exit:
pthread_mutex_unlock(&aperture->fmm_mutex);
return ret;
}