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

[SWDEV-475552/SWDEV-475351] Fix segfault TestComputePartitionReadWrite

Browse Source 
In order to check partition id's we must continue to check # of devices.
Since this fluctuates with partition updates
and there are drm minor limitations.

For the drm minor limitation of 64, user must remove other drivers
using PCIe space. You can see these by:
ls /sys/class/drm

Recommend: rmmod unneeded driver and reload amdgpu. In order to
ensure CPX can enumerate with all XCP (Graphic Cluster Partitions).

Change-Id: Ib663503f0b7264dce163f6ac2d50795fc8dc5eba
Signed-off-by: Charis Poag <Charis.Poag@amd.com>
This commit is contained in:
Charis Poag
2024-07-24 20:01:58 -05:00
committed by Maisam Arif
parent dd954be887
commit c11209f618
2 changed files with 83 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/rocm_smi_test/functional/computepartition_read_write.cc
+48
View File
@@ -157,6 +157,7 @@ static void checkPartitionIdChanges(
uint32_t dev, const std::string current_partition, bool isVerbose,
bool reinitialize) {
uint32_t max_loop = MAX_SPX_PARTITIONS;
uint32_t current_num_devices = 0;
// re-initialize to ensure new device ordering is followed
if (reinitialize) {
@@ -185,8 +186,19 @@ static void checkPartitionIdChanges(
}
}
}
rsmi_num_monitor_devices(&current_num_devices);
for (uint32_t i = dev; i < dev + max_loop; i++) {
if (dev+max_loop > current_num_devices) {
std::cout << "\t**Devices: " << dev << "; max_loop: " << max_loop
<< "; current_num_devices: " << current_num_devices << "\n";
std::cout << "\t**[WARNING] Detected max DRM minor limitation "
"(max of 64).\n\tPlease disable any other drivers taking up PCIe space"
"\n\t(ex. ast or other drivers -> "
"\"sudo rmmod amdgpu && sudo rmmod ast && sudo modprobe amdgpu\")."
"\n\tCPX may not enumerate properly.\n";
break;
}
uint32_t partition_id;
rsmi_status_t ret = rsmi_dev_partition_id_get(i, &partition_id);
std::cout << "\t** Checking Partition ID | Device: " << std::to_string(i)
@@ -419,7 +431,43 @@ void TestComputePartitionReadWrite::Run(void) {
*
*/
std::string final_partition_state = "UNKNOWN";
// loop through modes lower than original, but do not re-initialize this throws
// off device id numbers in tests
// (since we started with higher total # of devices)
for (int partition = static_cast<int>(RSMI_COMPUTE_PARTITION_SPX);
partition < static_cast<int>(mapStringToRSMIComputePartitionTypes.at(
std::string(orig_char_computePartition)));
partition++) {
if (std::string(orig_char_computePartition) == "SPX") {
break;
}
rsmi_compute_partition_type_t updatePartition
= static_cast<rsmi_compute_partition_type_t>(partition);
IF_VERB(STANDARD) {
std::cout << std::endl;
std::cout << "\t**"
<< "======== TEST RSMI_COMPUTE_PARTITION_"
<< computePartitionString(updatePartition)
<< " ===============" << std::endl;
}
ret = rsmi_dev_compute_partition_set(dv_ind, updatePartition);
IF_VERB(STANDARD) {
std::cout << "\t**"
<< "rsmi_dev_compute_partition_set(" << dv_ind
<< ", updatePartition): "
<< amd::smi::getRSMIStatusString(ret, false) << "\n"
<< "\t**New Partition (set): "
<< computePartitionString(updatePartition) << "\n";
}
ASSERT_TRUE((ret == RSMI_STATUS_SETTING_UNAVAILABLE)
|| (ret== RSMI_STATUS_PERMISSION)
|| (ret == RSMI_STATUS_SUCCESS)
|| ret == RSMI_STATUS_BUSY);
}
for (int partition = static_cast<int>(mapStringToRSMIComputePartitionTypes.at(
std::string(orig_char_computePartition)));
partition <= static_cast<int>(RSMI_COMPUTE_PARTITION_CPX);
partition++) {
rsmi_compute_partition_type_t updatePartition