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

[SWDEV-530385] show afids on each line of printout (#422)

Browse Source 
* show afids on each line of printout
* clean up afids and cper code
---------

Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Co-authored-by: Maisam Arif <Maisam.Arif@amd.com>

[ROCm/amdsmi commit: fab13c5b60]
This commit is contained in:
Saeed, Oosman
2025-06-02 17:22:10 -05:00
committed by GitHub
parent 3d75b7881a
commit 877c7b1bda
7 changed files with 107 additions and 197 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/amdsmi/src/amd_smi/amd_smi.cc
+1 -1
View File
@@ -4145,7 +4145,7 @@ amdsmi_status_t amdsmi_get_afids_from_cper(
return AMDSMI_STATUS_INVAL;
}
int i = 0;
uint32_t i = 0;
for(int afid: cper_decode(cper)) {
if(i < *num_afids) {
afids[i] = afid;
projects/amdsmi/src/amd_smi/amd_smi_cper.cc
+3 -5
View File
@@ -213,7 +213,6 @@ static void* cper_get_sec_desc_offset(const amdsmi_cper_hdr_t *hdr, int idx)
static void* cper_get_sec_offset(const amdsmi_cper_hdr_t *hdr, int idx)
{
struct cper_sec_desc *tmp_desc;
char *offset;
if (idx >= hdr->sec_cnt)
return 0;
@@ -313,14 +312,13 @@ static int cper_dump_nonstd_err(const struct cper_sec_nonstd_err *nonstd_err)
std::ostringstream ss;
struct cper_sec_nonstd_err_body *body;
char *offset;
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__ << "[AFIDS]\n~~~~NON STANDARD SECTION~~~\n";
ss << "[NonSTD SEC] Err Info Count = 0x" << std::hex << nonstd_err->hdr.valid_bits.err_info_cnt << "\n";
ss << "[NonSTD SEC] Err Context Count = 0x" << std::hex << nonstd_err->hdr.valid_bits.err_context_cnt << "\n";
if (nonstd_err->hdr.valid_bits.err_context_cnt != nonstd_err->hdr.valid_bits.err_context_cnt) {
if (nonstd_err->hdr.valid_bits.err_info_cnt != nonstd_err->hdr.valid_bits.err_context_cnt) {
ss << "~~~~Malformed Non Standard Section!~~~~\n\n";
goto exit;
}
@@ -554,8 +552,8 @@ std::vector<int> cper_decode(const amdsmi_cper_hdr_t *cper) {
}
else {
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__ << "[AFIDS] Unknown error type!!\n";
for(int i = 0; i < sizeof(sec_guid->b); ++i) {
ss << std::hex << static_cast<int>(sec_guid->b[i]) << ":";
for(size_t j = 0; j < sizeof(sec_guid->b); ++j) {
ss << std::hex << static_cast<int>(sec_guid->b[j]) << ":";
}
ss << "\n";
LOG_ERROR(ss);
projects/amdsmi/src/amd_smi/amd_smi_utils.cc
-87
View File
@@ -937,100 +937,13 @@ amdsmi_status_t smi_amdgpu_get_processor_handle_by_index(
return AMDSMI_STATUS_API_FAILED;
}
static std::vector<const amdsmi_cper_hdr_t *>
amdsmi_get_gpu_cper_headers(const char *buffer, size_t buffer_sz) {
std::ostringstream ss;
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__
<< "[CPER] buffer_sz: " << buffer_sz;
LOG_DEBUG(ss);
std::vector<const amdsmi_cper_hdr_t *> headers;
if(!buffer) {
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__
<< "[CPER] buffer is null";
LOG_ERROR(ss);
return headers;
}
static constexpr char cper_signature[] = "CPER";
static constexpr size_t cper_signature_size = sizeof(cper_signature) - 1;
for(size_t data_idx = 0;
buffer_sz >= cper_signature_size &&
data_idx < buffer_sz - cper_signature_size;
++data_idx) {
const amdsmi_cper_hdr_t *hdr = reinterpret_cast<const amdsmi_cper_hdr_t *>(
&buffer[data_idx]);
if(hdr->signature[0] != 'C' || hdr->signature[1] != 'P' ||
hdr->signature[2] != 'E' || hdr->signature[3] != 'R' ) {
continue;
}
if(hdr->signature_end != 0xFFFFFFFF) {
continue;
}
if(hdr->record_length > buffer_sz) {
continue;
}
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__
<< "[CPER] add header at data_idx: " << data_idx
<< ", sig: " << hdr->signature[0] << hdr->signature[1] << hdr->signature[2] << hdr->signature[3];
LOG_DEBUG(ss);
headers.emplace_back(hdr);
}
return headers;
}
struct CperFileCtx {
amdsmi_status_t status = AMDSMI_STATUS_FILE_ERROR;
std::unique_ptr<char[]> buffer;
long file_size = 0;
};
static auto amdsmi_read_cper_file(const std::string &filepath) -> CperFileCtx {
std::ostringstream ss;
CperFileCtx ctx;
ctx.status = AMDSMI_STATUS_FILE_ERROR;
ctx.file_size = 0;
struct stat file_stats;
if (stat(filepath.c_str(), &file_stats) == 0) {
if (!S_ISREG(file_stats.st_mode)) {
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__ << "[CPER] file is not a regular file: "
<< filepath << ", errno: " << errno << "): " << strerror(errno);
return ctx;
}
} else {
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__ << "[CPER] file does not exist: "
<< filepath << ", errno: " << errno << "): " << strerror(errno);
ctx.status = AMDSMI_STATUS_NOT_SUPPORTED;
return ctx;
}
ctx.file_size = file_stats.st_size;
ctx.buffer = std::make_unique<char[]>(ctx.file_size);
int file = open(filepath.c_str(), O_RDONLY);
if (file == -1) {
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__ << "[CPER] failed to open file: "
<< filepath << ", errno:()" << errno << "): " << strerror(errno);
LOG_ERROR(ss);
return ctx;
}
long bytes_read = read(file, ctx.buffer.get(), ctx.file_size);
if (bytes_read <= 0) {
ss << __PRETTY_FUNCTION__ << "\n:" << __LINE__
<< "[CPER] failed to read complete file, read only "
<< bytes_read << " of " << ctx.file_size << " bytes";
LOG_ERROR(ss);
return ctx;
}
close(file);
ctx.status = AMDSMI_STATUS_SUCCESS;
ctx.file_size = bytes_read;
return ctx;
}
void amdsmi_wait_for_user_input(void) {
for (;;) {
std::cout << "\n\t**Press any key to continue**" << std::endl;