static void
add_bdflt_to_case(fmd_hdl_t *hdl, char *label, const char *fltnm,
uint8_t board_cert, fmd_case_t *cp)
{
nvlist_t *memb_nvl, *flt;
memb_nvl = fru_by_label(hdl, label);
if (memb_nvl != NULL) {
flt = cmd_nvl_create_fault(hdl, fltnm, board_cert,
memb_nvl, memb_nvl, NULL);
flt = cmd_fault_add_location(hdl, flt, label);
if (flt != NULL) {
fmd_case_add_suspect(hdl, cp, flt);
}
nvlist_free(memb_nvl);
}
}
/*
* Solve a given ZFS case. This first checks to make sure the diagnosis is
* still valid, as well as cleaning up any pending timer associated with the
* case.
*/
static void
zfs_case_solve(fmd_hdl_t *hdl, zfs_case_t *zcp, const char *faultname,
boolean_t checkunusable)
{
nvlist_t *detector, *fault;
boolean_t serialize;
nvlist_t *fru = NULL;
fmd_hdl_debug(hdl, "solving fault '%s'", faultname);
/*
* Construct the detector from the case data. The detector is in the
* ZFS scheme, and is either the pool or the vdev, depending on whether
* this is a vdev or pool fault.
*/
detector = fmd_nvl_alloc(hdl, FMD_SLEEP);
(void) nvlist_add_uint8(detector, FM_VERSION, ZFS_SCHEME_VERSION0);
(void) nvlist_add_string(detector, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS);
(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_POOL,
zcp->zc_data.zc_pool_guid);
if (zcp->zc_data.zc_vdev_guid != 0) {
(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_VDEV,
zcp->zc_data.zc_vdev_guid);
}
fault = fmd_nvl_create_fault(hdl, faultname, 100, detector,
fru, detector);
fmd_case_add_suspect(hdl, zcp->zc_case, fault);
nvlist_free(fru);
fmd_case_solve(hdl, zcp->zc_case);
serialize = B_FALSE;
if (zcp->zc_data.zc_has_remove_timer) {
fmd_timer_remove(hdl, zcp->zc_remove_timer);
zcp->zc_data.zc_has_remove_timer = 0;
serialize = B_TRUE;
}
if (serialize)
zfs_case_serialize(hdl, zcp);
nvlist_free(detector);
}
/*
* 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);
}
/*
* Solve a given ZFS case. This first checks to make sure the diagnosis is
* still valid, as well as cleaning up any pending timer associated with the
* case.
*/
static void
zfs_case_solve(fmd_hdl_t *hdl, zfs_case_t *zcp, const char *faultname,
boolean_t checkunusable)
{
libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
nvlist_t *detector, *fault;
boolean_t serialize;
nvlist_t *fmri, *fru;
topo_hdl_t *thp;
int err;
/*
* Construct the detector from the case data. The detector is in the
* ZFS scheme, and is either the pool or the vdev, depending on whether
* this is a vdev or pool fault.
*/
detector = fmd_nvl_alloc(hdl, FMD_SLEEP);
(void) nvlist_add_uint8(detector, FM_VERSION, ZFS_SCHEME_VERSION0);
(void) nvlist_add_string(detector, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS);
(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_POOL,
zcp->zc_data.zc_pool_guid);
if (zcp->zc_data.zc_vdev_guid != 0) {
(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_VDEV,
zcp->zc_data.zc_vdev_guid);
}
/*
* We also want to make sure that the detector (pool or vdev) properly
* reflects the diagnosed state, when the fault corresponds to internal
* ZFS state (i.e. not checksum or I/O error-induced). Otherwise, a
* device which was unavailable early in boot (because the driver/file
* wasn't available) and is now healthy will be mis-diagnosed.
*/
if (!fmd_nvl_fmri_present(hdl, detector) ||
(checkunusable && !fmd_nvl_fmri_unusable(hdl, detector))) {
fmd_case_close(hdl, zcp->zc_case);
nvlist_free(detector);
return;
}
fru = NULL;
if (zcp->zc_fru != NULL &&
(thp = fmd_hdl_topo_hold(hdl, TOPO_VERSION)) != NULL) {
/*
* If the vdev had an associated FRU, then get the FRU nvlist
* from the topo handle and use that in the suspect list. We
* explicitly lookup the FRU because the fmri reported from the
* kernel may not have up to date details about the disk itself
* (serial, part, etc).
*/
if (topo_fmri_str2nvl(thp, zcp->zc_fru, &fmri, &err) == 0) {
/*
* If the disk is part of the system chassis, but the
* FRU indicates a different chassis ID than our
* current system, then ignore the error. This
* indicates that the device was part of another
* cluster head, and for obvious reasons cannot be
* imported on this system.
*/
if (libzfs_fru_notself(zhdl, zcp->zc_fru)) {
fmd_case_close(hdl, zcp->zc_case);
nvlist_free(fmri);
fmd_hdl_topo_rele(hdl, thp);
nvlist_free(detector);
return;
}
/*
* If the device is no longer present on the system, or
* topo_fmri_fru() fails for other reasons, then fall
* back to the fmri specified in the vdev.
*/
if (topo_fmri_fru(thp, fmri, &fru, &err) != 0)
fru = fmd_nvl_dup(hdl, fmri, FMD_SLEEP);
nvlist_free(fmri);
}
fmd_hdl_topo_rele(hdl, thp);
}
fault = fmd_nvl_create_fault(hdl, faultname, 100, detector,
fru, detector);
fmd_case_add_suspect(hdl, zcp->zc_case, fault);
nvlist_free(fru);
fmd_case_solve(hdl, zcp->zc_case);
serialize = B_FALSE;
if (zcp->zc_data.zc_has_remove_timer) {
fmd_timer_remove(hdl, zcp->zc_remove_timer);
zcp->zc_data.zc_has_remove_timer = 0;
serialize = B_TRUE;
}
if (serialize)
zfs_case_serialize(hdl, zcp);
nvlist_free(detector);
}
