Adjust how we read ECC block counter status
This change corresponds to kernel changes. Change-Id: Ibd977e8b3338349036cb16e55fb0b2c9c187726d
This commit is contained in:
1 vanhempi
a34832f11e
commit
aaecfd6fff
5 files changed
+68
-83
No files matched your search
@@ -98,6 +98,7 @@ void TestErrCntRead::Run(void) {
|
||||
std::cout <<
|
||||
"\t**Error Count Enabled Mask get is not supported on this machine"
|
||||
<< std::endl;
|
||||
continue;
|
||||
} else {
|
||||
CHK_ERR_ASRT(err)
|
||||
IF_VERB(STANDARD) {
|
||||
@@ -109,17 +110,11 @@ void TestErrCntRead::Run(void) {
|
||||
b <= RSMI_GPU_BLOCK_LAST; b = b*2) {
|
||||
err = rsmi_dev_ecc_status_get(i, static_cast<rsmi_gpu_block_t>(b),
|
||||
&err_state);
|
||||
if (err == RSMI_STATUS_NOT_SUPPORTED) {
|
||||
std::cout << "\t**Error Count Status for " <<
|
||||
GetBlockNameStr(static_cast<rsmi_gpu_block_t>(b)) <<
|
||||
": Not supported on this machine" << std::endl;
|
||||
} else {
|
||||
CHK_ERR_ASRT(err)
|
||||
IF_VERB(STANDARD) {
|
||||
std::cout << "\t**Error Count status for " <<
|
||||
GetBlockNameStr(static_cast<rsmi_gpu_block_t>(b)) <<
|
||||
" block: " << GetErrStateNameStr(err_state) << std::endl;
|
||||
}
|
||||
CHK_ERR_ASRT(err)
|
||||
IF_VERB(STANDARD) {
|
||||
std::cout << "\t**Error Count status for " <<
|
||||
GetBlockNameStr(static_cast<rsmi_gpu_block_t>(b)) <<
|
||||
" block: " << GetErrStateNameStr(err_state) << std::endl;
|
||||
}
|
||||
|
||||
err = rsmi_dev_ecc_count_get(i, static_cast<rsmi_gpu_block_t>(b), &ec);
|
||||
@@ -127,7 +122,7 @@ void TestErrCntRead::Run(void) {
|
||||
if (err == RSMI_STATUS_NOT_SUPPORTED) {
|
||||
std::cout << "\t**Error Count for " <<
|
||||
GetBlockNameStr(static_cast<rsmi_gpu_block_t>(b)) <<
|
||||
": Not supported on this machine" << std::endl;
|
||||
": Not supported for this device" << std::endl;
|
||||
} else {
|
||||
CHK_ERR_ASRT(err)
|
||||
IF_VERB(STANDARD) {
|
||||
|
||||
@@ -59,8 +59,19 @@ static const std::map<rsmi_gpu_block_t, const char *> kBlockNameMap = {
|
||||
{RSMI_GPU_BLOCK_UMC, "UMC"},
|
||||
{RSMI_GPU_BLOCK_SDMA, "SDMA"},
|
||||
{RSMI_GPU_BLOCK_GFX, "GFX"},
|
||||
{RSMI_GPU_BLOCK_MMHUB, "MMHUB"},
|
||||
{RSMI_GPU_BLOCK_ATHUB, "ATHUB"},
|
||||
{RSMI_GPU_BLOCK_PCIE_BIF, "PCIE_BIF"},
|
||||
{RSMI_GPU_BLOCK_HDP, "HDP"},
|
||||
{RSMI_GPU_BLOCK_XGMI_WAFL, "XGMI_WAFL"},
|
||||
{RSMI_GPU_BLOCK_DF, "DF"},
|
||||
{RSMI_GPU_BLOCK_SMN, "SMN"},
|
||||
{RSMI_GPU_BLOCK_SEM, "SEM"},
|
||||
{RSMI_GPU_BLOCK_MP0, "MP0"},
|
||||
{RSMI_GPU_BLOCK_MP1, "MP1"},
|
||||
{RSMI_GPU_BLOCK_FUSE, "FUSE"},
|
||||
};
|
||||
static_assert(RSMI_GPU_BLOCK_LAST == RSMI_GPU_BLOCK_GFX,
|
||||
static_assert(RSMI_GPU_BLOCK_LAST == RSMI_GPU_BLOCK_FUSE,
|
||||
"kBlockNameMap needs to be updated");
|
||||
|
||||
static const char * kRasErrStateStrings[] = {
|
||||
@@ -70,8 +81,8 @@ static const char * kRasErrStateStrings[] = {
|
||||
"Single, Correctable", // RSMI_RAS_ERR_STATE_SING_C
|
||||
"Multiple, Uncorrectable", // RSMI_RAS_ERR_STATE_MULT_UC
|
||||
"Poison" // RSMI_RAS_ERR_STATE_POISON
|
||||
"off", // RSMI_RAS_ERR_STATE_DISABLED
|
||||
"on", // RSMI_RAS_ERR_STATE_ENABLED
|
||||
"Off", // RSMI_RAS_ERR_STATE_DISABLED
|
||||
"On", // RSMI_RAS_ERR_STATE_ENABLED
|
||||
};
|
||||
static_assert(
|
||||
sizeof(kRasErrStateStrings)/sizeof(char *) == (RSMI_RAS_ERR_STATE_LAST + 1),
|
||||
|
||||
Reference in new issue
Block a user