/*
* For t5440, the memory channel goes like this:
* VF -> cpuboard -> D0 -> motherboard -> memboard -> D[1..3]
* If there is a dimm on the memory board, the memory board,
* motherboard, cpuboard, and dimms are in the suspect list.
* If there is no dimm on the memory board, the cpu board and
* the dimms are in the suspect list
* The board certainty = total board certainty / number of
* the faulty boards in the suspect list.
*/
void
cmd_branch_create_fault(fmd_hdl_t *hdl, cmd_branch_t *branch,
const char *fltnm, nvlist_t *asru)
{
nvlist_t *flt;
cmd_branch_memb_t *bm;
cmd_dimm_t *dimm;
int dimm_count = 0;
uint_t cert = 0;
uint_t board_cert = 0;
char *fruloc = NULL, *membd_label;
/* attach the dimms to the branch */
dimm_count = branch_dimmlist_create(hdl, branch);
if ((membd_label = mbd_label(hdl, branch, "MEM")) != NULL) {
board_cert = CMD_BOARDS_CERT / 3; /* CPU, MEM, MB */
/*
* Batoka with memory expansion. CPU expansion board will
* be added below. Add memory expansion board and motherboard
* FRUs here.
*/
add_bdflt_to_case(hdl, membd_label, fltnm, board_cert,
branch->branch_case.cc_cp);
fmd_hdl_strfree(hdl, membd_label);
add_bdflt_to_case(hdl, "MB", fltnm, board_cert,
branch->branch_case.cc_cp);
} else if ((membd_label = mbd_label(hdl, branch, "MR")) != NULL) {
board_cert = CMD_BOARDS_CERT / 2; /* MB, MR */
/*
* Maramba or similar platform with mezzanine board.
* Motherboard FRU will be added below. Add the mezzanine
* board here.
*/
add_bdflt_to_case(hdl, membd_label, fltnm, board_cert,
branch->branch_case.cc_cp);
fmd_hdl_strfree(hdl, membd_label);
} else {
board_cert = CMD_BOARDS_CERT; /* only MB or CPU */
}
/*
* The code which follows adds to the suspect list the FRU which
* contains the ereport 'detector'. This can be either a CPU
* expansion board (Batoka), or motherboard (Huron, Maramba, or
* derivative).
*/
fruloc = cmd_getfru_loc(hdl, asru);
flt = cmd_boardfru_create_fault(hdl, asru, fltnm, board_cert, fruloc);
if (flt != NULL)
fmd_case_add_suspect(hdl, branch->branch_case.cc_cp, flt);
if (dimm_count != 0)
cert = (100 - CMD_BOARDS_CERT) / dimm_count;
/* create dimm faults */
for (bm = cmd_list_next(&branch->branch_dimms); bm != NULL;
bm = cmd_list_next(bm)) {
dimm = bm->dimm;
if (dimm != NULL) {
dimm->dimm_flags |= CMD_MEM_F_FAULTING;
cmd_dimm_dirty(hdl, dimm);
flt = cmd_dimm_create_fault(hdl, dimm, fltnm, cert);
fmd_case_add_suspect(hdl, branch->branch_case.cc_cp,
flt);
}
}
if (fruloc != NULL)
fmd_hdl_strfree(hdl, fruloc);
}
cmd_dimm_t *
cmd_dimm_create(fmd_hdl_t *hdl, nvlist_t *asru)
{
cmd_dimm_t *dimm;
const char *unum;
nvlist_t *fmri;
size_t nserids = 0;
char **serids = NULL;
if (!fmd_nvl_fmri_present(hdl, asru)) {
fmd_hdl_debug(hdl, "dimm_lookup: discarding old ereport\n");
return (NULL);
}
if ((unum = cmd_fmri_get_unum(asru)) == NULL) {
CMD_STAT_BUMP(bad_mem_asru);
return (NULL);
}
#ifdef sun4v
if (nvlist_lookup_string_array(asru, FM_FMRI_HC_SERIAL_ID, &serids,
&nserids) != 0) {
fmd_hdl_debug(hdl, "sun4v mem: FMRI does not"
" have serial_ids\n");
CMD_STAT_BUMP(bad_mem_asru);
return (NULL);
}
#endif
fmri = cmd_mem_fmri_create(unum, serids, nserids);
if (fmd_nvl_fmri_expand(hdl, fmri) < 0) {
CMD_STAT_BUMP(bad_mem_asru);
nvlist_free(fmri);
return (NULL);
}
fmd_hdl_debug(hdl, "dimm_create: creating new DIMM %s\n", unum);
CMD_STAT_BUMP(dimm_creat);
dimm = fmd_hdl_zalloc(hdl, sizeof (cmd_dimm_t), FMD_SLEEP);
dimm->dimm_nodetype = CMD_NT_DIMM;
dimm->dimm_version = CMD_DIMM_VERSION;
cmd_bufname(dimm->dimm_bufname, sizeof (dimm->dimm_bufname), "dimm_%s",
unum);
cmd_fmri_init(hdl, &dimm->dimm_asru, fmri, "dimm_asru_%s", unum);
nvlist_free(fmri);
(void) nvlist_lookup_string(dimm->dimm_asru_nvl, FM_FMRI_MEM_UNUM,
(char **)&dimm->dimm_unum);
dimm_attach_to_bank(hdl, dimm);
cmd_mem_retirestat_create(hdl, &dimm->dimm_retstat, dimm->dimm_unum, 0,
CMD_DIMM_STAT_PREFIX);
cmd_list_append(&cmd.cmd_dimms, dimm);
cmd_dimm_dirty(hdl, dimm);
return (dimm);
}
