Turn on/off DAC capabilities as needed

Write access is required for some RSMI services. This change
temporarily permits write access so configuration can be done,
and then turns it off.

To help with this, the ScopedCapability struct is introduced to
provide scope limited access, helping to ensure a process is not
left with extra capability, should an exception occur.

Change-Id: I4978a1a688db935b8bfc27b3b537a0dd07959d3f
This commit is contained in:
Chris Freehill
2021-02-04 08:17:48 -06:00
parent a9d0e037b5
commit 6b5aeaaa23
8 changed files with 89 additions and 12 deletions
+3 -1
View File
@@ -163,6 +163,7 @@ set(RDC_LIB_SRC_LIST ${RDC_LIB_SRC_LIST} "${SRC_DIR}/rdc/src/RdcModuleMgrImpl.cc
set(RDC_LIB_SRC_LIST ${RDC_LIB_SRC_LIST} "${SRC_DIR}/rdc/src/RdcNotificationImpl.cc")
set(RDC_LIB_SRC_LIST ${RDC_LIB_SRC_LIST} "${SRC_DIR}/rdc/src/RsmiUtils.cc")
set(RDC_LIB_SRC_LIST ${RDC_LIB_SRC_LIST} "${COMMON_DIR}/rdc_fields_supported.cc")
set(RDC_LIB_SRC_LIST ${RDC_LIB_SRC_LIST} "${COMMON_DIR}/rdc_capabilities.cc")
set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${RDC_LIB_INC_DIR}/rdc_lib/impl/RdcEmbeddedHandler.h")
set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${RDC_LIB_INC_DIR}/rdc_lib/RdcMetricFetcher.h")
@@ -185,12 +186,13 @@ set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${RDC_LIB_INC_DIR}/rdc_lib/RdcNotifica
set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${RDC_LIB_INC_DIR}/rdc_lib/impl/RdcNotificationImpl.h")
set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${RDC_LIB_INC_DIR}/rdc_lib/impl/RsmiUtils.h")
set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${COMMON_DIR}/rdc_fields_supported.h")
set(RDC_LIB_INC_LIST ${RDC_LIB_INC_LIST} "${COMMON_DIR}/rdc_capabilities.h")
message("RDC_LIB_INC_LIST=${RDC_LIB_INC_LIST}")
link_directories(${RSMI_LIB_DIR} "${GRPC_ROOT}/lib" "${GRPC_ROOT}/lib64")
add_library(${RDC_LIB} SHARED ${RDC_LIB_SRC_LIST} ${RDC_LIB_INC_LIST})
target_link_libraries(${RDC_LIB} ${BOOTSTRAP_LIB} pthread rocm_smi64)
target_link_libraries(${RDC_LIB} ${BOOTSTRAP_LIB} pthread rocm_smi64 cap)
target_include_directories(${RDC_LIB} PRIVATE
"${PROJECT_SOURCE_DIR}"
"${PROJECT_SOURCE_DIR}/include"
+27
View File
@@ -33,6 +33,7 @@ THE SOFTWARE.
#include "rdc_lib/RdcLogger.h"
#include "rocm_smi/rocm_smi.h"
#include "rdc_lib/impl/RsmiUtils.h"
#include "common/rdc_capabilities.h"
namespace amd {
namespace rdc {
@@ -523,12 +524,32 @@ static rdc_status_t init_rsmi_counter(RdcFieldKey fk,
}
rsmi_event_type_t evt = rdc_evnt_2_rsmi_field.at(f);
// Temporarily get DAC capability
ScopedCapability sc(CAP_DAC_OVERRIDE, CAP_EFFECTIVE);
if (sc.error()) {
RDC_LOG(RDC_ERROR,
"Failed to acquire required capabilities. Errno " << sc.error());
return RDC_ST_PERM_ERROR;
}
ret = rsmi_dev_counter_create(dv_ind, evt, handle);
if (ret != RSMI_STATUS_SUCCESS) {
return Rsmi2RdcError(ret);
}
ret = rsmi_counter_control(*handle, RSMI_CNTR_CMD_START, nullptr);
// Release DAC capability
sc.Relinquish();
if (sc.error()) {
RDC_LOG(RDC_ERROR,
"Failed to relinquish capabilities. Errno " << sc.error());
return RDC_ST_PERM_ERROR;
}
return Rsmi2RdcError(ret);
}
@@ -561,6 +582,9 @@ rdc_status_t RdcMetricFetcherImpl::delete_rsmi_handle(RdcFieldKey fk) {
ret = rsmi_counter_control(h, RSMI_CNTR_CMD_STOP, nullptr);
if (ret != RSMI_STATUS_SUCCESS) {
rsmi_data_.erase(fk);
RDC_LOG(RDC_ERROR, "Error in stopping event counter: " <<
Rsmi2RdcError(ret));
return Rsmi2RdcError(ret);
}
@@ -590,6 +614,7 @@ rdc_status_t RdcMetricFetcherImpl::acquire_rsmi_handle(RdcFieldKey fk) {
result = init_rsmi_counter(fk, grp, &handle);
if (result != RDC_ST_OK) {
RDC_LOG(RDC_ERROR, "Failed to init RSMI counter. Return:" << result);
return result;
}
auto fsh = std::shared_ptr<FieldRSMIData>(new FieldRSMIData);
@@ -636,6 +661,8 @@ rdc_status_t RdcMetricFetcherImpl::acquire_rsmi_handle(RdcFieldKey fk) {
RDC_LOG(RDC_ERROR, "No event counters are available for " <<
field_id_to_descript.at(fk.second).enum_name << " event.");
} else if (ret != RDC_ST_OK) {
RDC_LOG(RDC_ERROR, "Error in getting event counter handle: " << ret);
}
return ret;
}
+22 -3
View File
@@ -34,6 +34,7 @@ THE SOFTWARE.
#include "rdc_lib/RdcLogger.h"
#include "rdc_lib/impl/RdcSmiLib.h"
#include "rocm_smi/rocm_smi.h"
#include "common/rdc_capabilities.h"
namespace amd {
namespace rdc {
@@ -95,9 +96,19 @@ RdcNotificationImpl::set_listen_events(const std::vector<RdcFieldKey> fk_arr) {
// No change to mask; nothing to be done
continue;
}
// Temporarily get DAC capability
ScopedCapability sc(CAP_DAC_OVERRIDE, CAP_EFFECTIVE);
if (sc.error()) {
RDC_LOG(RDC_ERROR,
"Failed to acquire required capabilities. Errno " << sc.error());
return RDC_ST_PERM_ERROR;
}
ret = rsmi_event_notification_init(it->first);
if (ret != RSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_INFO,
RDC_LOG(RDC_ERROR,
"rsmi_event_notification_init() returned " << ret << " for device " <<
it->first << ". " << std::endl <<
" Will not listen for events on this device");
@@ -105,6 +116,14 @@ RdcNotificationImpl::set_listen_events(const std::vector<RdcFieldKey> fk_arr) {
}
ret = rsmi_event_notification_mask_set(it->first, it->second);
// Release DAC capability
sc.Relinquish();
if (sc.error()) {
RDC_LOG(RDC_ERROR,
"Failed to relinquish capabilities. Errno " << sc.error());
return RDC_ST_PERM_ERROR;
}
if (ret == RSMI_STATUS_SUCCESS) {
gpu_evnt_notif_masks_[it->first] = it->second;
@@ -162,7 +181,7 @@ RdcNotificationImpl::stop_listening(uint32_t gpu_id) {
ret = rsmi_event_notification_mask_set(gpu_id, 0);
if (ret != RSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_INFO, "rsmi_event_notification_mask_set() returned " << ret
RDC_LOG(RDC_ERROR, "rsmi_event_notification_mask_set() returned " << ret
<< " for device " << gpu_id);
}
@@ -171,7 +190,7 @@ RdcNotificationImpl::stop_listening(uint32_t gpu_id) {
std::lock_guard<std::mutex> guard(notif_mutex_);
gpu_evnt_notif_masks_[gpu_id] = 0;
} else {
RDC_LOG(RDC_INFO, "rsmi_event_notification_stop() returned " << ret
RDC_LOG(RDC_ERROR, "rsmi_event_notification_stop() returned " << ret
<< " for device " << gpu_id);
}
return RDC_ST_OK;
+7 -1
View File
@@ -98,13 +98,19 @@ rdc_status_t RdcSmiLib::rdc_telemetry_fields_value_get(rdc_gpu_field_t* fields,
rdc_status_t RdcSmiLib::rdc_telemetry_fields_watch(rdc_gpu_field_t* fields,
uint32_t fields_count) {
rdc_status_t ret;
if (fields == nullptr) {
return RDC_ST_BAD_PARAMETER;
}
for (uint32_t i = 0; i < fields_count; i++) {
metric_fetcher_->acquire_rsmi_handle(
ret = metric_fetcher_->acquire_rsmi_handle(
{fields[i].gpu_index, fields[i].field_id});
if (ret != RDC_ST_OK) {
RDC_LOG(RDC_ERROR,
"Failed to acquire rocm_smi handle for field.");
}
}
RDC_LOG(RDC_DEBUG, "acquire " << fields_count
<< " field handles from rocm_smi_lib");