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

Fix RAS change

Browse Source 
RAS formatting changed, so get it to handle both types of sysfs output
until it's normalized
Change-Id: I56f2a2495af8ff4d01011bc614283376afb9ad0a
This commit is contained in:
Kent Russell
2019-08-08 09:00:22 -04:00
parent 0da1599c4f
commit a34832f11e
3 changed files with 14 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/rocm_smi/rocm_smi.h
+2 -1
View File
@@ -383,8 +383,9 @@ typedef enum {
RSMI_RAS_ERR_STATE_MULT_UC, //!< Multiple uncorrectable errors RSMI_RAS_ERR_STATE_MULT_UC, //!< Multiple uncorrectable errors
RSMI_RAS_ERR_STATE_POISON, //!< Firmware detected error and isolated RSMI_RAS_ERR_STATE_POISON, //!< Firmware detected error and isolated
//!< page. Treat as uncorrectable. //!< page. Treat as uncorrectable.
RSMI_RAS_ERR_STATE_ENABLED, //!< ECC is enabled
RSMI_RAS_ERR_STATE_LAST = RSMI_RAS_ERR_STATE_POISON, RSMI_RAS_ERR_STATE_LAST = RSMI_RAS_ERR_STATE_ENABLED,
RSMI_RAS_ERR_STATE_INVALID = 0xFFFFFFFF RSMI_RAS_ERR_STATE_INVALID = 0xFFFFFFFF
} rsmi_ras_err_state_t; } rsmi_ras_err_state_t;
src/rocm_smi.cc
+7 -3
View File
@@ -532,8 +532,10 @@ static const std::map<std::string, rsmi_ras_err_state_t> kRocmSMIStateMap = {
{"single_correctable", RSMI_RAS_ERR_STATE_SING_C}, {"single_correctable", RSMI_RAS_ERR_STATE_SING_C},
{"multi_uncorrectable", RSMI_RAS_ERR_STATE_MULT_UC}, {"multi_uncorrectable", RSMI_RAS_ERR_STATE_MULT_UC},
{"poison", RSMI_RAS_ERR_STATE_POISON}, {"poison", RSMI_RAS_ERR_STATE_POISON},
{"off", RSMI_RAS_ERR_STATE_DISABLED},
{"on", RSMI_RAS_ERR_STATE_ENABLED},
}; };
static_assert(RSMI_RAS_ERR_STATE_LAST == RSMI_RAS_ERR_STATE_POISON, static_assert(RSMI_RAS_ERR_STATE_LAST == RSMI_RAS_ERR_STATE_ENABLED,
"rsmi_gpu_block_t and/or above name map need to be updated" "rsmi_gpu_block_t and/or above name map need to be updated"
" and then this assert"); " and then this assert");
@@ -562,6 +564,7 @@ rsmi_status_t rsmi_dev_ecc_status_get(uint32_t dv_ind, rsmi_gpu_block_t block,
std::string blk_line; std::string blk_line;
std::string search_str = kRocmSMIBlockMap.at(block); std::string search_str = kRocmSMIBlockMap.at(block);
std::string sysfs_junk = " ras feature mask:";
std::string state_str; std::string state_str;
search_str += ":"; search_str += ":";
@@ -570,8 +573,9 @@ rsmi_status_t rsmi_dev_ecc_status_get(uint32_t dv_ind, rsmi_gpu_block_t block,
std::istringstream fs1(val_vec[i]); std::istringstream fs1(val_vec[i]);
fs1 >> blk_line; fs1 >> blk_line;
if (blk_line == search_str || blk_line == kRocmSMIBlockMap.at(block)) {
if (blk_line == search_str) { if (blk_line.back() != ':')
fs1.ignore(sysfs_junk.length(), ':');
fs1 >> state_str; fs1 >> state_str;
assert(kRocmSMIStateMap.count(state_str)); assert(kRocmSMIStateMap.count(state_str));
*state = kRocmSMIStateMap.at(state_str); *state = kRocmSMIStateMap.at(state_str);
tests/rocm_smi_test/test_common.cc
+5 -1
View File
@@ -70,6 +70,8 @@ static const char * kRasErrStateStrings[] = {
"Single, Correctable", // RSMI_RAS_ERR_STATE_SING_C "Single, Correctable", // RSMI_RAS_ERR_STATE_SING_C
"Multiple, Uncorrectable", // RSMI_RAS_ERR_STATE_MULT_UC "Multiple, Uncorrectable", // RSMI_RAS_ERR_STATE_MULT_UC
"Poison" // RSMI_RAS_ERR_STATE_POISON "Poison" // RSMI_RAS_ERR_STATE_POISON
"off", // RSMI_RAS_ERR_STATE_DISABLED
"on", // RSMI_RAS_ERR_STATE_ENABLED
}; };
static_assert( static_assert(
sizeof(kRasErrStateStrings)/sizeof(char *) == (RSMI_RAS_ERR_STATE_LAST + 1), sizeof(kRasErrStateStrings)/sizeof(char *) == (RSMI_RAS_ERR_STATE_LAST + 1),
@@ -89,8 +91,10 @@ static const std::map<rsmi_ras_err_state_t, const char *> kErrStateNameMap = {
kRasErrStateStrings[RSMI_RAS_ERR_STATE_MULT_UC]}, kRasErrStateStrings[RSMI_RAS_ERR_STATE_MULT_UC]},
{RSMI_RAS_ERR_STATE_POISON, {RSMI_RAS_ERR_STATE_POISON,
kRasErrStateStrings[RSMI_RAS_ERR_STATE_POISON]}, kRasErrStateStrings[RSMI_RAS_ERR_STATE_POISON]},
{RSMI_RAS_ERR_STATE_ENABLED,
kRasErrStateStrings[RSMI_RAS_ERR_STATE_ENABLED]},
}; };
static_assert(RSMI_RAS_ERR_STATE_LAST == RSMI_RAS_ERR_STATE_POISON, static_assert(RSMI_RAS_ERR_STATE_LAST == RSMI_RAS_ERR_STATE_ENABLED,
"kErrStateNameMap needs to be updated"); "kErrStateNameMap needs to be updated");
static const struct option long_options[] = { static const struct option long_options[] = {