Skip to content

fix(amdsmi): initialize Device metrics fields to avoid uninitialized reads #8182

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
marifamd merged 6 commits into from
Aug 28, 2026
Merged

fix(amdsmi): initialize Device metrics fields to avoid uninitialized reads #8182

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5cdbff1
[#3:TRIAGE][AMDSMI]amdsmi_topo_get_p2p_status use-after-free and amds…
SakaSitharammurthy Jul 7, 2026
8b3e7db
fix(amdsmi): complete uninitialized-read cleanup and add p2p null-arg…
marifamd Jul 20, 2026
8a16bad
fix(amdsmi):Updated profile_config & resource_index values
SakaSitharammurthy Jul 23, 2026
4903d36
Addressed review comments
SakaSitharammurthy Aug 4, 2026
6660838
Merge branch 'develop' into users/ssaka/fix_uninitialized_read
SakaSitharammurthy Aug 13, 2026
dbf7d73
Merge branch 'develop' into users/ssaka/fix_uninitialized_read
SakaSitharammurthy Aug 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 8 additions & 1 deletion projects/amdsmi/rocm_smi/src/rocm_smi_device.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -691,7 +691,14 @@ static const std::map<const char*, dev_depends_t> kDevFuncDependsMap = {
}

Device::Device(std::string p, RocmSMI_env_vars const* e)
: monitor_(nullptr), path_(p), env_(e), evt_notif_anon_fd_(-1), m_gpu_metrics_header{0, 0, 0} {
: monitor_(nullptr),
path_(p),
env_(e),
evt_notif_anon_fd_(-1),
Comment thread
SakaSitharammurthy marked this conversation as resolved.
Resolved
m_gpu_metrics_header{0, 0, 0},
m_gpu_metrics_updated_timestamp(0),
m_device_id(0),
m_partition_id(0) {
Comment thread
SakaSitharammurthy marked this conversation as resolved.
Resolved
#ifndef DEBUG
env_ = nullptr;
#endif
Expand Down
26 changes: 21 additions & 5 deletions projects/amdsmi/src/amd_smi/amd_smi.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3039,6 +3039,9 @@ amdsmi_status_t amdsmi_topo_get_p2p_status(amdsmi_processor_handle processor_han
amdsmi_processor_handle processor_handle_dst,
amdsmi_link_type_t* type, amdsmi_p2p_capability_t* cap) {
AMDSMI_CHECK_INIT();
if (type == nullptr || cap == nullptr) {
return AMDSMI_STATUS_INVAL;
}

amd::smi::AMDSmiGPUDevice* src_device = nullptr;
amd::smi::AMDSmiGPUDevice* dst_device = nullptr;
Expand Down Expand Up @@ -3180,12 +3183,15 @@ amdsmi_status_t amdsmi_get_gpu_memory_partition_config(amdsmi_processor_handle p
return AMDSMI_STATUS_INVAL;
}

*config = {};

// initialization for devices which do not support partitions
amdsmi_nps_caps_t flags;
flags.nps_flags.nps1_cap = 0;
flags.nps_flags.nps2_cap = 0;
flags.nps_flags.nps4_cap = 0;
flags.nps_flags.nps8_cap = 0;
flags.nps_cap_mask = 0;
config->partition_caps = flags;
config->mp_mode = AMDSMI_MEMORY_PARTITION_UNKNOWN;
// TODO(amdsmi_team): Will BM/guest VMs have numa ranges?
Expand Down Expand Up @@ -3219,7 +3225,8 @@ amdsmi_status_t amdsmi_get_gpu_memory_partition_config(amdsmi_processor_handle p
auto status_mem_caps = rsmi_wrapper(rsmi_dev_memory_partition_capabilities_get, processor_handle,
0, memory_caps, kLenCapsSize);
ss << __PRETTY_FUNCTION__ << " | rsmi_dev_memory_partition_capabilities_get Returning: "
<< smi_amdgpu_get_status_string(status, false) << " | Type: memory_partition_capabilities"
<< smi_amdgpu_get_status_string(status_mem_caps, false)
<< " | Type: memory_partition_capabilities"
<< " | Data: " << memory_caps;
LOG_DEBUG(ss);
std::string memory_caps_str = "N/A";
Expand Down Expand Up @@ -3282,6 +3289,7 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile_config(
flags.nps_flags.nps2_cap = 0;
flags.nps_flags.nps4_cap = 0;
flags.nps_flags.nps8_cap = 0;
flags.nps_cap_mask = 0;

ss << __PRETTY_FUNCTION__ << " | 1";
// std::cout << ss.str() << std::endl;
Expand All @@ -3293,7 +3301,7 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile_config(
// ex. SPX, DPX, QPX, CPX
std::string accelerator_caps_str = "N/A";
constexpr uint32_t kLenXCPConfigSize = 30;
char supported_xcp_configs[kLenXCPConfigSize];
char supported_xcp_configs[kLenXCPConfigSize] = {};
bool use_xcp_config = false;
return_status = rsmi_wrapper(rsmi_dev_compute_partition_supported_xcp_configs_get,
processor_handle, 0, supported_xcp_configs, kLenXCPConfigSize);
Expand Down Expand Up @@ -3456,6 +3464,12 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile_config(
rsmi_accelerator_partition_resource_type_t type =
static_cast<rsmi_accelerator_partition_resource_type_t>(r);
rsmi_accelerator_partition_resource_profile_t profile;
profile.partition_resource = 0;
profile.num_partitions_share_resource = 0;
profile_config->resource_profiles[resource_index].profile_index = 0;
profile_config->resource_profiles[resource_index].resource_type = AMDSMI_ACCELERATOR_MAX;
profile_config->resource_profiles[resource_index].partition_resource = 0;
profile_config->resource_profiles[resource_index].num_partitions_share_resource = 0;
Comment thread
SakaSitharammurthy marked this conversation as resolved.
Resolved
status = rsmi_wrapper(rsmi_dev_compute_partition_resource_profile_get, processor_handle, 0,
&type, &profile);
if (status == AMDSMI_STATUS_SUCCESS) {
Expand Down Expand Up @@ -3665,6 +3679,7 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile(
flags.nps_flags.nps2_cap = 0;
flags.nps_flags.nps4_cap = 0;
flags.nps_flags.nps8_cap = 0;
flags.nps_cap_mask = 0;
profile->memory_caps = flags;

// TODO(amdsmi_team): add resources here ^
Expand All @@ -3683,13 +3698,14 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile(
// ex. SPX = 0, DPX = 1, QPX = 2, CPX = 3; other devices may have different values
std::string accelerator_capabilities = "N/A";
constexpr uint32_t kLenXCPConfigSize = 30;
char supported_xcp_configs[kLenXCPConfigSize];
char supported_xcp_configs[kLenXCPConfigSize] = {};
bool use_xcp_config = false;
status = rsmi_wrapper(rsmi_dev_compute_partition_supported_xcp_configs_get, processor_handle, 0,
supported_xcp_configs, kLenXCPConfigSize);
if (status == AMDSMI_STATUS_SUCCESS) {
accelerator_capabilities.clear();
accelerator_capabilities = std::string(supported_xcp_configs);
accelerator_capabilities = amd::smi::trimAllWhiteSpace(accelerator_capabilities);
use_xcp_config = true;
}

Expand Down Expand Up @@ -3728,7 +3744,7 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile(
<< current_partition << "|";
LOG_DEBUG(ss);
current_partition_str = current_partition;
if (status == AMDSMI_STATUS_SUCCESS) {
if (compute_status == AMDSMI_STATUS_SUCCESS) {
// 1) get profile index from
// /sys/class/drm/../device/compute_partition_config/supported_xcp_configs
if (current_partition_str == "SPX" || current_partition_str == "DPX" ||
Expand Down Expand Up @@ -3814,7 +3830,7 @@ amdsmi_status_t amdsmi_get_gpu_accelerator_partition_profile(

// Add memory partition capabilities here
constexpr uint32_t kLenCapsSize = 30;
char memory_caps[kLenCapsSize];
char memory_caps[kLenCapsSize] = {};
status = rsmi_wrapper(rsmi_dev_memory_partition_capabilities_get, processor_handle, 0,
memory_caps, kLenCapsSize);
ss << __PRETTY_FUNCTION__ << " | rsmi_dev_memory_partition_capabilities_get Returning: "
Expand Down
Loading