[SWDEV-523349/SWDEV-527257] Fix Rdci Config (#161)

Change-Id: Iae21ea8061205f186086a3ed59c6259ddeb1dbe7

Signed-off-by: adapryor <Adam.pryor@amd.com>
This commit is contained in:
Pryor, Adam
2025-04-28 11:57:51 -05:00
کامیت شده توسط GitHub
والد 9edacdeac4
کامیت 2db6ddea69
10فایلهای تغییر یافته به همراه66 افزوده شده و 56 حذف شده
@@ -44,8 +44,7 @@ void RdcConfigSettingsImpl::monitorSettings() {
rdc_status_t rdc_status;
rdc_group_info_t rdc_group_info = {};
amdsmi_power_cap_info_t cap_info = {};
amdsmi_dev_perf_level_t perf_info = {};
uint32_t od;
amdsmi_frequencies_t freqs = {};
uint64_t cached_value;
while (true) {
@@ -103,19 +102,18 @@ void RdcConfigSettingsImpl::monitorSettings() {
}
// Mem clock
status = amdsmi_get_gpu_overdrive_level(processor_handle, &od);
status = amdsmi_get_clk_freq(processor_handle, AMDSMI_CLK_TYPE_MEM, &freqs);
if (status != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(
RDC_ERROR,
"RdcConfigSettingsImpl::monitorSettings(); amdsmi_get_gpu_overdrive_level failed: "
<< status);
"RdcConfigSettingsImpl::monitorSettings(); amdsmi_get_clk_freq failed: " << status);
continue;
}
auto mem_clk_it = cached_settings.find(RDC_CFG_MEMORY_CLOCK_LIMIT);
if (mem_clk_it != cached_settings.end()) {
cached_value = mem_clk_it->second.target_value;
if (od == cached_value) {
if (freqs.frequency[freqs.current] == cached_value) {
status = amdsmi_set_gpu_clk_limit(processor_handle, AMDSMI_CLK_TYPE_MEM,
CLK_LIMIT_MAX, cached_value);
if (status != AMDSMI_STATUS_SUCCESS) {
@@ -129,18 +127,18 @@ void RdcConfigSettingsImpl::monitorSettings() {
}
// GFX clock
status = amdsmi_get_gpu_perf_level(processor_handle, &perf_info);
status = amdsmi_get_clk_freq(processor_handle, AMDSMI_CLK_TYPE_GFX, &freqs);
if (status != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_ERROR,
"RdcConfigSettingsImpl::monitorSettings(); amdsmi_get_gpu_perf_level failed: "
<< status);
RDC_LOG(
RDC_ERROR,
"RdcConfigSettingsImpl::monitorSettings(); amdsmi_get_clk_freq failed: " << status);
continue;
}
auto gfx_clk_it = cached_settings.find(RDC_CFG_GFX_CLOCK_LIMIT);
if (gfx_clk_it != cached_settings.end()) {
cached_value = gfx_clk_it->second.target_value;
if (od == cached_value) {
if (freqs.frequency[freqs.current] == cached_value) {
status = amdsmi_set_gpu_clk_limit(processor_handle, AMDSMI_CLK_TYPE_GFX,
CLK_LIMIT_MAX, cached_value);
if (status != AMDSMI_STATUS_SUCCESS) {
@@ -313,31 +311,40 @@ rdc_status_t RdcConfigSettingsImpl::rdc_config_clear(rdc_gpu_group_t group_id) {
// Reset GFX clock limit if it was set
if (group_iter->second.find(RDC_CFG_GFX_CLOCK_LIMIT) != group_iter->second.end()) {
amdsmi_dev_perf_level_t perf_info = {};
amd_ret = amdsmi_get_gpu_perf_level(processor_handle, &perf_info);
if (amd_ret == AMDSMI_STATUS_SUCCESS && perf_info != AMDSMI_DEV_PERF_LEVEL_AUTO) {
amd_ret = amdsmi_set_gpu_clk_limit(processor_handle, AMDSMI_CLK_TYPE_GFX, CLK_LIMIT_MAX,
AMDSMI_DEV_PERF_LEVEL_AUTO);
if (amd_ret != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_ERROR,
"RdcConfigSettingsImpl::rdc_config_clear: Failed to reset GFX clock limit : "
<< amd_ret);
break;
amdsmi_frequencies_t freqs = {};
amd_ret = amdsmi_get_clk_freq(processor_handle, AMDSMI_CLK_TYPE_GFX, &freqs);
if (amd_ret == AMDSMI_STATUS_SUCCESS) {
uint64_t curr = freqs.frequency[freqs.current];
uint64_t maxf = freqs.frequency[freqs.num_supported - 1];
if (curr != maxf) {
amd_ret = amdsmi_set_gpu_clk_limit(processor_handle, AMDSMI_CLK_TYPE_GFX, CLK_LIMIT_MAX,
AMDSMI_DEV_PERF_LEVEL_AUTO);
if (amd_ret != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_ERROR,
"RdcConfigSettingsImpl::rdc_config_clear: Failed to reset GFX clock limit : "
<< amd_ret);
break;
}
}
}
}
// Reset memory clock limit if it was set
if (group_iter->second.find(RDC_CFG_MEMORY_CLOCK_LIMIT) != group_iter->second.end()) {
uint32_t od = 0;
amd_ret = amdsmi_get_gpu_overdrive_level(processor_handle, &od);
if (amd_ret == AMDSMI_STATUS_SUCCESS && od != 0) {
amd_ret = amdsmi_set_gpu_clk_limit(processor_handle, AMDSMI_CLK_TYPE_MEM, CLK_LIMIT_MAX, 0);
if (amd_ret != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_ERROR,
"RdcConfigSettingsImpl::rdc_config_clear: Failed to reset memory clock limit:"
<< amd_ret);
break;
amdsmi_frequencies_t freqs = {};
amd_ret = amdsmi_get_clk_freq(processor_handle, AMDSMI_CLK_TYPE_MEM, &freqs);
if (amd_ret == AMDSMI_STATUS_SUCCESS) {
uint64_t curr = freqs.frequency[freqs.current];
uint64_t maxf = freqs.frequency[freqs.num_supported - 1];
if (curr != maxf) {
amd_ret =
amdsmi_set_gpu_clk_limit(processor_handle, AMDSMI_CLK_TYPE_MEM, CLK_LIMIT_MAX, 0);
if (amd_ret != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_ERROR,
"RdcConfigSettingsImpl::rdc_config_clear: Failed to reset memory clock limit:"
<< amd_ret);
break;
}
}
}
}
@@ -210,7 +210,7 @@ rdc_status_t RdcEmbeddedHandler::rdc_device_get_component_version(
return RDC_ST_BAD_PARAMETER;
}
if (component == RDC_AMDMSI_COMPONENT) {
if (component == RDC_AMDSMI_COMPONENT) {
amdsmi_status_t ret;
amdsmi_version_t ver = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, nullptr};
@@ -219,7 +219,7 @@ rdc_status_t RdcEmbeddedHandler::rdc_device_get_component_version(
if (ret != AMDSMI_STATUS_SUCCESS) {
RDC_LOG(RDC_ERROR, "Failed to obtain the version of the server's amd-smi library. reason: "
<< (ret == AMDSMI_STATUS_INVAL ? "Invalid parameters" : "unknown"));
return RDC_ST_MSI_ERROR;
return RDC_ST_SMI_ERROR;
}
strncpy_with_null(p_rdc_compv->version, ver.build, RDC_MAX_VERSION_STR_LENGTH);
@@ -1125,7 +1125,7 @@ rdc_status_t RdcMetricFetcherImpl::fetch_smi_field(uint32_t gpu_index, rdc_field
<< value->value.str << ", latency " << latency);
}
return value->status == AMDSMI_STATUS_SUCCESS ? RDC_ST_OK : RDC_ST_MSI_ERROR;
return value->status == AMDSMI_STATUS_SUCCESS ? RDC_ST_OK : RDC_ST_SMI_ERROR;
}
std::shared_ptr<FieldSMIData> RdcMetricFetcherImpl::get_smi_data(RdcFieldKey key) {
@@ -82,7 +82,7 @@ rdc_status_t RdcPolicyImpl::rdc_policy_set(rdc_gpu_group_t group_id, rdc_policy_
gpu_index = group_info.entity_ids[i];
status = metric_fetcher_->fetch_smi_field(gpu_index, RDC_FI_GPU_PAGE_RETRIED, &value);
if (status == RDC_ST_MSI_ERROR) return RDC_ST_NOT_SUPPORTED;
if (status == RDC_ST_SMI_ERROR) return RDC_ST_NOT_SUPPORTED;
}
}
@@ -212,7 +212,7 @@ rdc_status_t RdcSmiDiagnosticImpl::check_smi_topo_info(uint32_t gpu_index[RDC_MA
err_info += " fail";
strncpy_with_null(result->details.msg, err_info.c_str(), MAX_DIAG_MSG_LENGTH);
strncpy_with_null(result->info, err_info.c_str(), MAX_DIAG_MSG_LENGTH);
return RDC_ST_MSI_ERROR;
return RDC_ST_SMI_ERROR;
}
info += std::to_string(i) + "=>";