/**********************************************************************
* Send Trap 64 on all new ports.
**********************************************************************/
static void state_mgr_report_new_ports(IN osm_sm_t * sm)
{
ib_gid_t port_gid;
ib_mad_notice_attr_t notice;
ib_api_status_t status;
ib_net64_t port_guid;
cl_map_item_t *p_next;
osm_port_t *p_port;
uint16_t min_lid_ho;
uint16_t max_lid_ho;
OSM_LOG_ENTER(sm->p_log);
CL_PLOCK_ACQUIRE(sm->p_lock);
p_next = cl_qmap_head(&sm->p_subn->port_guid_tbl);
while (p_next != cl_qmap_end(&sm->p_subn->port_guid_tbl)) {
p_port = (osm_port_t *) p_next;
p_next = cl_qmap_next(p_next);
if (!p_port->is_new)
continue;
port_guid = osm_port_get_guid(p_port);
/* issue a notice - trap 64 */
/* details of the notice */
notice.generic_type = 0x83; /* is generic subn mgt type */
ib_notice_set_prod_type_ho(¬ice, 4); /* A Class Manager generator */
/* endport becomes reachable */
notice.g_or_v.generic.trap_num = CL_HTON16(64);
/* The sm_base_lid is saved in network order already. */
notice.issuer_lid = sm->p_subn->sm_base_lid;
/* following C14-72.1.1 and table 119 p739 */
/* we need to provide the GID */
port_gid.unicast.prefix = sm->p_subn->opt.subnet_prefix;
port_gid.unicast.interface_id = port_guid;
memcpy(&(notice.data_details.ntc_64_67.gid), &(port_gid),
sizeof(ib_gid_t));
/* According to page 653 - the issuer gid in this case of trap
* is the SM gid, since the SM is the initiator of this trap. */
notice.issuer_gid.unicast.prefix =
sm->p_subn->opt.subnet_prefix;
notice.issuer_gid.unicast.interface_id =
sm->p_subn->sm_port_guid;
status = osm_report_notice(sm->p_log, sm->p_subn, ¬ice);
if (status != IB_SUCCESS)
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3318: "
"Error sending trap reports on GUID:0x%016"
PRIx64 " (%s)\n", port_gid.unicast.interface_id,
ib_get_err_str(status));
osm_port_get_lid_range_ho(p_port, &min_lid_ho, &max_lid_ho);
OSM_LOG(sm->p_log, OSM_LOG_INFO,
"Discovered new port with GUID:0x%016" PRIx64
" LID range [%u,%u] of node: %s\n",
cl_ntoh64(port_gid.unicast.interface_id),
min_lid_ho, max_lid_ho,
p_port->p_node ? p_port->p_node->
print_desc : "UNKNOWN");
p_port->is_new = 0;
}
CL_PLOCK_RELEASE(sm->p_lock);
OSM_LOG_EXIT(sm->p_log);
}
/****f* opensm: SM/__osm_sm_mad_ctrl_process_trap
* NAME
* __osm_sm_mad_ctrl_process_trap
*
* DESCRIPTION
* This function handles method Trap() for received MADs.
*
* SYNOPSIS
*/
static void
__osm_sm_mad_ctrl_process_trap(IN osm_sm_mad_ctrl_t * const p_ctrl,
IN osm_madw_t * p_madw)
{
ib_smp_t *p_smp;
cl_status_t status;
cl_disp_msgid_t msg_id = CL_DISP_MSGID_NONE;
OSM_LOG_ENTER(p_ctrl->p_log);
p_smp = osm_madw_get_smp_ptr(p_madw);
/* Make sure OpenSM is master. If not - then we should not process the trap */
if (p_ctrl->p_subn->sm_state != IB_SMINFO_STATE_MASTER) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_DEBUG,
"Received trap but OpenSM is not in MASTER state. "
"Dropping mad\n");
osm_mad_pool_put(p_ctrl->p_mad_pool, p_madw);
goto Exit;
}
/*
Note that attr_id (like the rest of the MAD) is in
network byte order.
*/
switch (p_smp->attr_id) {
case IB_MAD_ATTR_NOTICE:
msg_id = OSM_MSG_MAD_NOTICE;
break;
default:
cl_atomic_inc(&p_ctrl->p_stats->qp0_mads_rcvd_unknown);
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3109: "
"Unsupported attribute = 0x%X\n",
cl_ntoh16(p_smp->attr_id));
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_ERROR);
break;
}
if (msg_id == CL_DISP_MSGID_NONE) {
/*
There is an unknown MAD attribute type for which there is
no recipient. Simply retire the MAD here.
*/
osm_mad_pool_put(p_ctrl->p_mad_pool, p_madw);
goto Exit;
}
/*
Post this MAD to the dispatcher for asynchronous
processing by the appropriate controller.
*/
OSM_LOG(p_ctrl->p_log, OSM_LOG_DEBUG, "Posting Dispatcher message %s\n",
osm_get_disp_msg_str(msg_id));
status = cl_disp_post(p_ctrl->h_disp, msg_id, p_madw,
__osm_sm_mad_ctrl_disp_done_callback, p_ctrl);
if (status != CL_SUCCESS) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3110: "
"Dispatcher post message failed (%s)\n",
CL_STATUS_MSG(status));
goto Exit;
}
Exit:
OSM_LOG_EXIT(p_ctrl->p_log);
}
/****f* opensm: SM/__osm_sm_mad_ctrl_send_err_cb
* NAME
* __osm_sm_mad_ctrl_send_err_cb
*
* DESCRIPTION
* This is the callback from the transport layer for send errors
* on MADs that were expecting a response.
*
* SYNOPSIS
*/
static void
__osm_sm_mad_ctrl_send_err_cb(IN void *bind_context, IN osm_madw_t * p_madw)
{
osm_sm_mad_ctrl_t *p_ctrl = (osm_sm_mad_ctrl_t *) bind_context;
ib_api_status_t status;
ib_smp_t *p_smp;
OSM_LOG_ENTER(p_ctrl->p_log);
CL_ASSERT(p_madw);
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3113: "
"MAD completed in error (%s)\n",
ib_get_err_str(p_madw->status));
/*
If this was a SubnSet MAD, then this error might indicate a problem
in configuring the subnet. In this case - need to mark that there was
such a problem. The subnet will not be up, and the next sweep should
be a heavy sweep as well.
*/
p_smp = osm_madw_get_smp_ptr(p_madw);
if (p_smp->method == IB_MAD_METHOD_SET &&
(p_smp->attr_id == IB_MAD_ATTR_PORT_INFO ||
p_smp->attr_id == IB_MAD_ATTR_MCAST_FWD_TBL ||
p_smp->attr_id == IB_MAD_ATTR_SWITCH_INFO ||
p_smp->attr_id == IB_MAD_ATTR_LIN_FWD_TBL)) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3119: "
"Set method failed\n");
p_ctrl->p_subn->subnet_initialization_error = TRUE;
}
/*
Since we did not get any response we suspect the DR path
used for the target port.
Find it and replace it with an alternate path.
This is true only if the destination lid is not 0xFFFF, since
then we are aiming for a specific path and not specific destination
lid.
*/
/* For now - do not add the alternate dr path to the release */
#if 0
if (p_madw->mad_addr.dest_lid != 0xFFFF) {
osm_physp_t *p_physp =
osm_get_physp_by_mad_addr(p_ctrl->p_log,
p_ctrl->p_subn,
&(p_madw->mad_addr));
if (!p_physp) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3114: "
"Failed to find the corresponding phys port\n");
} else {
osm_physp_replace_dr_path_with_alternate_dr_path
(p_ctrl->p_log, p_ctrl->p_subn, p_physp,
p_madw->h_bind);
}
}
#endif
/*
An error occurred. No response was received to a request MAD.
Retire the original request MAD.
*/
osm_dump_dr_smp(p_ctrl->p_log, osm_madw_get_smp_ptr(p_madw),
OSM_LOG_ERROR);
__osm_sm_mad_ctrl_update_wire_stats(p_ctrl);
if (osm_madw_get_err_msg(p_madw) != CL_DISP_MSGID_NONE) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_DEBUG,
"Posting Dispatcher message %s\n",
osm_get_disp_msg_str(osm_madw_get_err_msg(p_madw)));
status = cl_disp_post(p_ctrl->h_disp,
osm_madw_get_err_msg(p_madw),
p_madw,
__osm_sm_mad_ctrl_disp_done_callback,
p_ctrl);
if (status != CL_SUCCESS)
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3115: "
"Dispatcher post message failed (%s)\n",
CL_STATUS_MSG(status));
} else
/*
No error message was provided, just retire the MAD.
*/
__osm_sm_mad_ctrl_retire_trans_mad(p_ctrl, p_madw);
OSM_LOG_EXIT(p_ctrl->p_log);
}
static void sa_slvl_by_comp_mask(IN osm_sa_t * sa, IN const osm_port_t * p_port,
osm_slvl_search_ctxt_t * p_ctxt)
{
const ib_slvl_table_record_t *p_rcvd_rec;
ib_net64_t comp_mask;
const osm_physp_t *p_out_physp, *p_in_physp;
uint8_t in_port_num, out_port_num;
uint8_t num_ports;
uint8_t in_port_start, in_port_end;
uint8_t out_port_start, out_port_end;
const osm_physp_t *p_req_physp;
OSM_LOG_ENTER(sa->p_log);
p_rcvd_rec = p_ctxt->p_rcvd_rec;
comp_mask = p_ctxt->comp_mask;
num_ports = osm_node_get_num_physp(p_port->p_node);
in_port_start = 0;
in_port_end = num_ports - 1;
out_port_start = 0;
out_port_end = num_ports - 1;
p_req_physp = p_ctxt->p_req_physp;
if (p_port->p_node->node_info.node_type != IB_NODE_TYPE_SWITCH) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Using Physical Default Port Number: 0x%X (for End Node)\n",
p_port->p_physp->port_num);
p_out_physp = p_port->p_physp;
/* check that the p_out_physp and the p_req_physp share a pkey */
if (osm_physp_share_pkey(sa->p_log, p_req_physp, p_out_physp))
sa_slvl_create(sa, p_out_physp, p_ctxt, 0);
} else {
if (comp_mask & IB_SLVL_COMPMASK_OUT_PORT)
out_port_start = out_port_end =
p_rcvd_rec->out_port_num;
if (comp_mask & IB_SLVL_COMPMASK_IN_PORT)
in_port_start = in_port_end = p_rcvd_rec->in_port_num;
for (out_port_num = out_port_start;
out_port_num <= out_port_end; out_port_num++) {
p_out_physp =
osm_node_get_physp_ptr(p_port->p_node,
out_port_num);
if (!p_out_physp)
continue;
for (in_port_num = in_port_start;
in_port_num <= in_port_end; in_port_num++) {
#if 0
if (out_port_num && out_port_num == in_port_num)
continue;
#endif
p_in_physp =
osm_node_get_physp_ptr(p_port->p_node,
in_port_num);
if (!p_in_physp)
continue;
/* if the requester and the p_out_physp don't share a pkey -
continue */
if (!osm_physp_share_pkey
(sa->p_log, p_req_physp, p_out_physp))
continue;
sa_slvl_create(sa, p_out_physp, p_ctxt,
in_port_num);
}
}
}
OSM_LOG_EXIT(sa->p_log);
}
static int link_mgr_set_physp_pi(osm_sm_t * sm, IN osm_physp_t * p_physp,
IN uint8_t port_state)
{
uint8_t payload[IB_SMP_DATA_SIZE], payload2[IB_SMP_DATA_SIZE];
ib_port_info_t *p_pi = (ib_port_info_t *) payload;
ib_mlnx_ext_port_info_t *p_epi = (ib_mlnx_ext_port_info_t *) payload2;
const ib_port_info_t *p_old_pi;
const ib_mlnx_ext_port_info_t *p_old_epi;
osm_madw_context_t context;
osm_node_t *p_node;
ib_api_status_t status;
uint8_t port_num, mtu, op_vls, smsl = OSM_DEFAULT_SL;
boolean_t esp0 = FALSE, send_set = FALSE, send_set2 = FALSE;
osm_physp_t *p_remote_physp, *physp0;
int qdr_change = 0, fdr10_change = 0;
int ret = 0;
ib_net32_t attr_mod, cap_mask;
OSM_LOG_ENTER(sm->p_log);
p_node = osm_physp_get_node_ptr(p_physp);
p_old_pi = &p_physp->port_info;
port_num = osm_physp_get_port_num(p_physp);
if (port_num == 0) {
/*
CAs don't have a port 0, and for switch port 0,
we need to check if this is enhanced or base port 0.
For base port 0 the following parameters are not valid
(IBA 1.2.1 p.830 table 146).
*/
if (!p_node->sw) {
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 4201: "
"Cannot find switch by guid: 0x%" PRIx64 "\n",
cl_ntoh64(p_node->node_info.node_guid));
goto Exit;
}
if (ib_switch_info_is_enhanced_port0(&p_node->sw->switch_info)
== FALSE) {
/* Even for base port 0 we might have to set smsl
(if we are using lash routing) */
smsl = link_mgr_get_smsl(sm, p_physp);
if (smsl != ib_port_info_get_master_smsl(p_old_pi)) {
send_set = TRUE;
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Setting SMSL to %d on port 0 GUID 0x%016"
PRIx64 "\n", smsl,
cl_ntoh64(osm_physp_get_port_guid
(p_physp)));
} else {
/* This means the switch doesn't support
enhanced port 0 and we don't need to
change SMSL. Can skip it. */
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Skipping port 0, GUID 0x%016" PRIx64
"\n",
cl_ntoh64(osm_physp_get_port_guid
(p_physp)));
goto Exit;
}
} else
esp0 = TRUE;
}
memcpy(payload, p_old_pi, sizeof(ib_port_info_t));
/*
Should never write back a value that is bigger then 3 in
the PortPhysicalState field - so can not simply copy!
Actually we want to write there:
port physical state - no change,
link down default state = polling
port state - as requested.
*/
p_pi->state_info2 = 0x02;
ib_port_info_set_port_state(p_pi, port_state);
/* Check whether this is base port0 smsl handling only */
if (port_num == 0 && esp0 == FALSE) {
ib_port_info_set_master_smsl(p_pi, smsl);
goto Send;
}
/*
PAST THIS POINT WE ARE HANDLING EITHER A NON PORT 0 OR ENHANCED PORT 0
*/
if (ib_port_info_get_link_down_def_state(p_pi) !=
ib_port_info_get_link_down_def_state(p_old_pi))
send_set = TRUE;
/* didn't get PortInfo before */
if (!ib_port_info_get_port_state(p_old_pi))
send_set = TRUE;
/* we only change port fields if we do not change state */
if (port_state == IB_LINK_NO_CHANGE) {
/* The following fields are relevant only for CA port, router, or Enh. SP0 */
if (osm_node_get_type(p_node) != IB_NODE_TYPE_SWITCH ||
port_num == 0) {
p_pi->m_key = sm->p_subn->opt.m_key;
if (memcmp(&p_pi->m_key, &p_old_pi->m_key,
sizeof(p_pi->m_key)))
send_set = TRUE;
p_pi->subnet_prefix = sm->p_subn->opt.subnet_prefix;
if (memcmp(&p_pi->subnet_prefix,
&p_old_pi->subnet_prefix,
sizeof(p_pi->subnet_prefix)))
send_set = TRUE;
p_pi->base_lid = osm_physp_get_base_lid(p_physp);
if (memcmp(&p_pi->base_lid, &p_old_pi->base_lid,
sizeof(p_pi->base_lid)))
send_set = TRUE;
/* we are initializing the ports with our local sm_base_lid */
p_pi->master_sm_base_lid = sm->p_subn->sm_base_lid;
if (memcmp(&p_pi->master_sm_base_lid,
&p_old_pi->master_sm_base_lid,
sizeof(p_pi->master_sm_base_lid)))
send_set = TRUE;
smsl = link_mgr_get_smsl(sm, p_physp);
if (smsl != ib_port_info_get_master_smsl(p_old_pi)) {
ib_port_info_set_master_smsl(p_pi, smsl);
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Setting SMSL to %d on GUID 0x%016"
PRIx64 ", port %d\n", smsl,
cl_ntoh64(osm_physp_get_port_guid
(p_physp)), port_num);
send_set = TRUE;
}
p_pi->m_key_lease_period =
sm->p_subn->opt.m_key_lease_period;
if (memcmp(&p_pi->m_key_lease_period,
&p_old_pi->m_key_lease_period,
sizeof(p_pi->m_key_lease_period)))
send_set = TRUE;
/* M_KeyProtectBits are currently always zero */
p_pi->mkey_lmc = 0;
if (esp0 == FALSE || sm->p_subn->opt.lmc_esp0)
ib_port_info_set_lmc(p_pi, sm->p_subn->opt.lmc);
if (ib_port_info_get_lmc(p_old_pi) !=
ib_port_info_get_lmc(p_pi) ||
ib_port_info_get_mpb(p_old_pi) !=
ib_port_info_get_mpb(p_pi))
send_set = TRUE;
ib_port_info_set_timeout(p_pi,
sm->p_subn->opt.
subnet_timeout);
if (ib_port_info_get_timeout(p_pi) !=
ib_port_info_get_timeout(p_old_pi))
send_set = TRUE;
}
/*
Several timeout mechanisms:
*/
p_remote_physp = osm_physp_get_remote(p_physp);
if (port_num != 0 && p_remote_physp) {
if (osm_node_get_type(osm_physp_get_node_ptr(p_physp))
== IB_NODE_TYPE_ROUTER) {
ib_port_info_set_hoq_lifetime(p_pi,
sm->p_subn->
opt.
leaf_head_of_queue_lifetime);
} else
if (osm_node_get_type
(osm_physp_get_node_ptr(p_physp)) ==
IB_NODE_TYPE_SWITCH) {
/* Is remote end CA or router (a leaf port) ? */
if (osm_node_get_type
(osm_physp_get_node_ptr(p_remote_physp)) !=
IB_NODE_TYPE_SWITCH) {
ib_port_info_set_hoq_lifetime(p_pi,
sm->
p_subn->
opt.
leaf_head_of_queue_lifetime);
ib_port_info_set_vl_stall_count(p_pi,
sm->
p_subn->
opt.
leaf_vl_stall_count);
} else {
ib_port_info_set_hoq_lifetime(p_pi,
sm->
p_subn->
opt.
head_of_queue_lifetime);
ib_port_info_set_vl_stall_count(p_pi,
sm->
p_subn->
opt.
vl_stall_count);
}
}
if (ib_port_info_get_hoq_lifetime(p_pi) !=
ib_port_info_get_hoq_lifetime(p_old_pi) ||
ib_port_info_get_vl_stall_count(p_pi) !=
ib_port_info_get_vl_stall_count(p_old_pi))
send_set = TRUE;
}
ib_port_info_set_phy_and_overrun_err_thd(p_pi,
sm->p_subn->opt.
local_phy_errors_threshold,
sm->p_subn->opt.
overrun_errors_threshold);
if (memcmp(&p_pi->error_threshold, &p_old_pi->error_threshold,
sizeof(p_pi->error_threshold)))
send_set = TRUE;
/*
Set the easy common parameters for all port types,
then determine the neighbor MTU.
*/
p_pi->link_width_enabled = p_old_pi->link_width_supported;
if (memcmp(&p_pi->link_width_enabled,
&p_old_pi->link_width_enabled,
sizeof(p_pi->link_width_enabled)))
send_set = TRUE;
if (sm->p_subn->opt.force_link_speed &&
(sm->p_subn->opt.force_link_speed != 15 ||
ib_port_info_get_link_speed_enabled(p_pi) !=
ib_port_info_get_link_speed_sup(p_pi))) {
ib_port_info_set_link_speed_enabled(p_pi,
sm->p_subn->opt.
force_link_speed);
if (memcmp(&p_pi->link_speed, &p_old_pi->link_speed,
sizeof(p_pi->link_speed))) {
send_set = TRUE;
/* Determine whether QDR in LSE is being changed */
if ((ib_port_info_get_link_speed_enabled(p_pi) &
IB_LINK_SPEED_ACTIVE_10 &&
!(ib_port_info_get_link_speed_enabled(p_old_pi) &
IB_LINK_SPEED_ACTIVE_10)) ||
((!(ib_port_info_get_link_speed_enabled(p_pi) &
IB_LINK_SPEED_ACTIVE_10) &&
ib_port_info_get_link_speed_enabled(p_old_pi) &
IB_LINK_SPEED_ACTIVE_10)))
qdr_change = 1;
}
}
if (sm->p_subn->opt.fdr10 &&
p_physp->ext_port_info.link_speed_supported & FDR10) {
if (sm->p_subn->opt.fdr10 == 1) { /* enable */
if (!(p_physp->ext_port_info.link_speed_enabled & FDR10))
fdr10_change = 1;
} else { /* disable */
if (p_physp->ext_port_info.link_speed_enabled & FDR10)
fdr10_change = 1;
}
if (fdr10_change) {
p_old_epi = &p_physp->ext_port_info;
memcpy(payload2, p_old_epi,
sizeof(ib_mlnx_ext_port_info_t));
p_epi->state_change_enable = 0x01;
if (sm->p_subn->opt.fdr10 == 1)
p_epi->link_speed_enabled = FDR10;
else
p_epi->link_speed_enabled = 0;
send_set2 = TRUE;
}
}
if (osm_node_get_type(p_physp->p_node) == IB_NODE_TYPE_SWITCH) {
physp0 = osm_node_get_physp_ptr(p_physp->p_node, 0);
cap_mask = physp0->port_info.capability_mask;
} else
cap_mask = p_pi->capability_mask;
if (!(cap_mask & IB_PORT_CAP_HAS_EXT_SPEEDS))
qdr_change = 0;
/* Do peer ports support extended link speeds ? */
if (port_num != 0 && p_remote_physp) {
osm_physp_t *rphysp0;
ib_net32_t rem_cap_mask;
if (osm_node_get_type(p_remote_physp->p_node) ==
IB_NODE_TYPE_SWITCH) {
rphysp0 = osm_node_get_physp_ptr(p_remote_physp->p_node, 0);
rem_cap_mask = rphysp0->port_info.capability_mask;
} else
rem_cap_mask = p_remote_physp->port_info.capability_mask;
if (cap_mask & IB_PORT_CAP_HAS_EXT_SPEEDS &&
rem_cap_mask & IB_PORT_CAP_HAS_EXT_SPEEDS) {
if (sm->p_subn->opt.force_link_speed_ext &&
(sm->p_subn->opt.force_link_speed_ext != IB_LINK_SPEED_EXT_SET_LSES ||
p_pi->link_speed_ext_enabled !=
ib_port_info_get_link_speed_sup(p_pi))) {
p_pi->link_speed_ext_enabled = sm->p_subn->opt.force_link_speed_ext;
if (memcmp(&p_pi->link_speed_ext_enabled,
&p_old_pi->link_speed_ext_enabled,
sizeof(p_pi->link_speed_ext_enabled)))
send_set = TRUE;
}
}
}
/* calc new op_vls and mtu */
op_vls =
osm_physp_calc_link_op_vls(sm->p_log, sm->p_subn, p_physp);
mtu = osm_physp_calc_link_mtu(sm->p_log, p_physp);
ib_port_info_set_neighbor_mtu(p_pi, mtu);
if (ib_port_info_get_neighbor_mtu(p_pi) !=
ib_port_info_get_neighbor_mtu(p_old_pi))
send_set = TRUE;
ib_port_info_set_op_vls(p_pi, op_vls);
if (ib_port_info_get_op_vls(p_pi) !=
ib_port_info_get_op_vls(p_old_pi))
send_set = TRUE;
/* provide the vl_high_limit from the qos mgr */
if (sm->p_subn->opt.qos &&
p_physp->vl_high_limit != p_old_pi->vl_high_limit) {
send_set = TRUE;
p_pi->vl_high_limit = p_physp->vl_high_limit;
}
}
Send:
if (port_state != IB_LINK_NO_CHANGE &&
port_state != ib_port_info_get_port_state(p_old_pi)) {
send_set = TRUE;
if (port_state == IB_LINK_ACTIVE)
context.pi_context.active_transition = TRUE;
else
context.pi_context.active_transition = FALSE;
}
context.pi_context.node_guid = osm_node_get_node_guid(p_node);
context.pi_context.port_guid = osm_physp_get_port_guid(p_physp);
context.pi_context.set_method = TRUE;
context.pi_context.light_sweep = FALSE;
/* We need to send the PortInfoSet request with the new sm_lid
in the following cases:
1. There is a change in the values (send_set == TRUE)
2. This is a switch external port (so it wasn't handled yet by
osm_lid_mgr) and first_time_master_sweep flag on the subnet is TRUE,
which means the SM just became master, and it then needs to send at
PortInfoSet to every port.
*/
if (osm_node_get_type(p_node) == IB_NODE_TYPE_SWITCH && port_num
&& sm->p_subn->first_time_master_sweep == TRUE)
send_set = TRUE;
if (!send_set)
goto Exit;
attr_mod = cl_hton32(port_num);
if (qdr_change)
attr_mod |= cl_hton32(1 << 31); /* AM SMSupportExtendedSpeeds */
status = osm_req_set(sm, osm_physp_get_dr_path_ptr(p_physp),
payload, sizeof(payload), IB_MAD_ATTR_PORT_INFO,
attr_mod, CL_DISP_MSGID_NONE, &context);
if (status)
ret = -1;
if (send_set2) {
status = osm_req_set(sm, osm_physp_get_dr_path_ptr(p_physp),
payload2, sizeof(payload2),
IB_MAD_ATTR_MLNX_EXTENDED_PORT_INFO,
cl_hton32(port_num),
CL_DISP_MSGID_NONE, &context);
if (status)
ret = -1;
}
Exit:
OSM_LOG_EXIT(sm->p_log);
return ret;
}
osm_bind_handle_t
osmv_bind_sa(IN osm_vendor_t * const p_vend,
IN osm_mad_pool_t * const p_mad_pool, IN ib_net64_t port_guid)
{
osm_bind_info_t bind_info;
osm_log_t *p_log = p_vend->p_log;
osmv_sa_bind_info_t *p_sa_bind_info;
cl_status_t cl_status;
OSM_LOG_ENTER(p_log);
OSM_LOG(p_log, OSM_LOG_DEBUG,
"Binding to port 0x%" PRIx64 "\n", cl_ntoh64(port_guid));
bind_info.port_guid = port_guid;
bind_info.mad_class = IB_MCLASS_SUBN_ADM;
bind_info.class_version = 2;
bind_info.is_responder = FALSE;
bind_info.is_trap_processor = FALSE;
bind_info.is_report_processor = FALSE;
bind_info.send_q_size = OSM_SM_DEFAULT_QP1_RCV_SIZE;
bind_info.recv_q_size = OSM_SM_DEFAULT_QP1_SEND_SIZE;
/* allocate the new sa bind info */
p_sa_bind_info =
(osmv_sa_bind_info_t *) malloc(sizeof(osmv_sa_bind_info_t));
if (!p_sa_bind_info) {
OSM_LOG(p_log, OSM_LOG_ERROR, "ERR 5505: "
"Failed to allocate new bind structure\n");
p_sa_bind_info = OSM_BIND_INVALID_HANDLE;
goto Exit;
}
/* store some important context */
p_sa_bind_info->p_log = p_log;
p_sa_bind_info->p_mad_pool = p_mad_pool;
p_sa_bind_info->p_vendor = p_vend;
/* Bind to the lower level */
p_sa_bind_info->h_bind = osm_vendor_bind(p_vend, &bind_info, p_mad_pool, __osmv_sa_mad_rcv_cb, __osmv_sa_mad_err_cb, p_sa_bind_info); /* context provided to CBs */
if (p_sa_bind_info->h_bind == OSM_BIND_INVALID_HANDLE) {
free(p_sa_bind_info);
p_sa_bind_info = OSM_BIND_INVALID_HANDLE;
OSM_LOG(p_log, OSM_LOG_ERROR, "ERR 5506: "
"Failed to bind to vendor GSI\n");
goto Exit;
}
/* update time umad_port is initilized now */
p_sa_bind_info->last_lids_update_sec = time(NULL);
/* initialize the sync_event */
cl_event_construct(&p_sa_bind_info->sync_event);
cl_status = cl_event_init(&p_sa_bind_info->sync_event, TRUE);
if (cl_status != CL_SUCCESS) {
OSM_LOG(p_log, OSM_LOG_ERROR, "ERR 5508: "
"cl_init_event failed: %s\n", ib_get_err_str(cl_status));
free(p_sa_bind_info);
p_sa_bind_info = OSM_BIND_INVALID_HANDLE;
}
Exit:
OSM_LOG_EXIT(p_log);
return (p_sa_bind_info);
}
/*
* Query the SA based on the user's request.
*/
ib_api_status_t
osmv_query_sa(IN osm_bind_handle_t h_bind,
IN const osmv_query_req_t * const p_query_req)
{
osmv_sa_bind_info_t *p_bind = (osmv_sa_bind_info_t *) h_bind;
osmv_sa_mad_data_t sa_mad_data;
osmv_user_query_t *p_user_query;
ib_service_record_t svc_rec;
ib_node_record_t node_rec;
ib_portinfo_record_t port_info;
ib_path_rec_t path_rec;
#ifdef DUAL_SIDED_RMPP
ib_multipath_rec_t multipath_rec;
osmv_multipath_req_t *p_mpr_req;
int i, j;
#endif
ib_class_port_info_t class_port_info;
osm_log_t *p_log = p_bind->p_log;
ib_api_status_t status;
OSM_LOG_ENTER(p_log);
/* Set the request information. */
sa_mad_data.method = IB_MAD_METHOD_GETTABLE;
sa_mad_data.attr_mod = 0;
/* Set the MAD attributes and component mask correctly. */
switch (p_query_req->query_type) {
case OSMV_QUERY_USER_DEFINED:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 USER_DEFINED\n");
p_user_query = (osmv_user_query_t *) p_query_req->p_query_input;
if (p_user_query->method)
sa_mad_data.method = p_user_query->method;
sa_mad_data.attr_offset = p_user_query->attr_offset;
sa_mad_data.attr_id = p_user_query->attr_id;
sa_mad_data.attr_mod = p_user_query->attr_mod;
sa_mad_data.comp_mask = p_user_query->comp_mask;
sa_mad_data.p_attr = p_user_query->p_attr;
break;
case OSMV_QUERY_ALL_SVC_RECS:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 SVC_REC_BY_NAME\n");
sa_mad_data.attr_id = IB_MAD_ATTR_SERVICE_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_service_record_t));
sa_mad_data.comp_mask = 0;
sa_mad_data.p_attr = &svc_rec;
break;
case OSMV_QUERY_SVC_REC_BY_NAME:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 SVC_REC_BY_NAME\n");
sa_mad_data.method = IB_MAD_METHOD_GET;
sa_mad_data.attr_id = IB_MAD_ATTR_SERVICE_RECORD;
sa_mad_data.comp_mask = IB_SR_COMPMASK_SNAME;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_service_record_t));
sa_mad_data.p_attr = &svc_rec;
memcpy(svc_rec.service_name, p_query_req->p_query_input,
sizeof(ib_svc_name_t));
break;
case OSMV_QUERY_SVC_REC_BY_ID:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 SVC_REC_BY_ID\n");
sa_mad_data.attr_id = IB_MAD_ATTR_SERVICE_RECORD;
sa_mad_data.comp_mask = IB_SR_COMPMASK_SID;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_service_record_t));
sa_mad_data.p_attr = &svc_rec;
svc_rec.service_id =
*(ib_net64_t *) (p_query_req->p_query_input);
break;
case OSMV_QUERY_CLASS_PORT_INFO:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 CLASS_PORT_INFO\n");
sa_mad_data.method = IB_MAD_METHOD_GET;
sa_mad_data.attr_id = IB_MAD_ATTR_CLASS_PORT_INFO;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_class_port_info_t));
sa_mad_data.comp_mask = 0;
sa_mad_data.p_attr = &class_port_info;
break;
case OSMV_QUERY_NODE_REC_BY_NODE_GUID:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 NODE_REC_BY_NODE_GUID\n");
sa_mad_data.attr_id = IB_MAD_ATTR_NODE_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_node_record_t));
sa_mad_data.comp_mask = IB_NR_COMPMASK_NODEGUID;
sa_mad_data.p_attr = &node_rec;
node_rec.node_info.node_guid =
*(ib_net64_t *) (p_query_req->p_query_input);
break;
case OSMV_QUERY_PORT_REC_BY_LID:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 PORT_REC_BY_LID\n");
sa_mad_data.attr_id = IB_MAD_ATTR_PORTINFO_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_portinfo_record_t));
sa_mad_data.comp_mask = IB_PIR_COMPMASK_LID;
sa_mad_data.p_attr = &port_info;
port_info.lid = *(ib_net16_t *) (p_query_req->p_query_input);
break;
case OSMV_QUERY_PORT_REC_BY_LID_AND_NUM:
sa_mad_data.method = IB_MAD_METHOD_GET;
p_user_query = (osmv_user_query_t *) p_query_req->p_query_input;
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 PORT_REC_BY_LID_AND_NUM\n");
sa_mad_data.attr_id = IB_MAD_ATTR_PORTINFO_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_portinfo_record_t));
sa_mad_data.comp_mask =
IB_PIR_COMPMASK_LID | IB_PIR_COMPMASK_PORTNUM;
sa_mad_data.p_attr = p_user_query->p_attr;
break;
case OSMV_QUERY_VLARB_BY_LID_PORT_BLOCK:
sa_mad_data.method = IB_MAD_METHOD_GET;
p_user_query = (osmv_user_query_t *) p_query_req->p_query_input;
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 OSMV_QUERY_VLARB_BY_LID_PORT_BLOCK\n");
sa_mad_data.attr_id = IB_MAD_ATTR_VLARB_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_vl_arb_table_record_t));
sa_mad_data.comp_mask =
IB_VLA_COMPMASK_LID | IB_VLA_COMPMASK_OUT_PORT |
IB_VLA_COMPMASK_BLOCK;
sa_mad_data.p_attr = p_user_query->p_attr;
break;
case OSMV_QUERY_SLVL_BY_LID_AND_PORTS:
sa_mad_data.method = IB_MAD_METHOD_GET;
p_user_query = (osmv_user_query_t *) p_query_req->p_query_input;
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 OSMV_QUERY_VLARB_BY_LID_PORT_BLOCK\n");
sa_mad_data.attr_id = IB_MAD_ATTR_SLVL_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_slvl_table_record_t));
sa_mad_data.comp_mask =
IB_SLVL_COMPMASK_LID | IB_SLVL_COMPMASK_OUT_PORT |
IB_SLVL_COMPMASK_IN_PORT;
sa_mad_data.p_attr = p_user_query->p_attr;
break;
case OSMV_QUERY_PATH_REC_BY_PORT_GUIDS:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 PATH_REC_BY_PORT_GUIDS\n");
memset(&path_rec, 0, sizeof(ib_path_rec_t));
sa_mad_data.attr_id = IB_MAD_ATTR_PATH_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_path_rec_t));
sa_mad_data.comp_mask =
(IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID);
sa_mad_data.p_attr = &path_rec;
ib_gid_set_default(&path_rec.dgid,
((osmv_guid_pair_t *) (p_query_req->
p_query_input))->
dest_guid);
ib_gid_set_default(&path_rec.sgid,
((osmv_guid_pair_t *) (p_query_req->
p_query_input))->
src_guid);
break;
case OSMV_QUERY_PATH_REC_BY_GIDS:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 PATH_REC_BY_GIDS\n");
memset(&path_rec, 0, sizeof(ib_path_rec_t));
sa_mad_data.attr_id = IB_MAD_ATTR_PATH_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_path_rec_t));
sa_mad_data.comp_mask =
(IB_PR_COMPMASK_DGID | IB_PR_COMPMASK_SGID);
sa_mad_data.p_attr = &path_rec;
memcpy(&path_rec.dgid,
&((osmv_gid_pair_t *) (p_query_req->p_query_input))->
dest_gid, sizeof(ib_gid_t));
memcpy(&path_rec.sgid,
&((osmv_gid_pair_t *) (p_query_req->p_query_input))->
src_gid, sizeof(ib_gid_t));
break;
case OSMV_QUERY_PATH_REC_BY_LIDS:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 PATH_REC_BY_LIDS\n");
memset(&path_rec, 0, sizeof(ib_path_rec_t));
sa_mad_data.method = IB_MAD_METHOD_GET;
sa_mad_data.attr_id = IB_MAD_ATTR_PATH_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_path_rec_t));
sa_mad_data.comp_mask =
(IB_PR_COMPMASK_DLID | IB_PR_COMPMASK_SLID);
sa_mad_data.p_attr = &path_rec;
path_rec.dlid =
((osmv_lid_pair_t *) (p_query_req->p_query_input))->
dest_lid;
path_rec.slid =
((osmv_lid_pair_t *) (p_query_req->p_query_input))->src_lid;
break;
case OSMV_QUERY_UD_MULTICAST_SET:
sa_mad_data.method = IB_MAD_METHOD_SET;
p_user_query = (osmv_user_query_t *) p_query_req->p_query_input;
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 OSMV_QUERY_UD_MULTICAST_SET\n");
sa_mad_data.attr_id = IB_MAD_ATTR_MCMEMBER_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_member_rec_t));
sa_mad_data.comp_mask = p_user_query->comp_mask;
sa_mad_data.p_attr = p_user_query->p_attr;
break;
case OSMV_QUERY_UD_MULTICAST_DELETE:
sa_mad_data.method = IB_MAD_METHOD_DELETE;
p_user_query = (osmv_user_query_t *) p_query_req->p_query_input;
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 OSMV_QUERY_UD_MULTICAST_DELETE\n");
sa_mad_data.attr_id = IB_MAD_ATTR_MCMEMBER_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_member_rec_t));
sa_mad_data.comp_mask = p_user_query->comp_mask;
sa_mad_data.p_attr = p_user_query->p_attr;
break;
#ifdef DUAL_SIDED_RMPP
case OSMV_QUERY_MULTIPATH_REC:
OSM_LOG(p_log, OSM_LOG_DEBUG, "DBG:001 MULTIPATH_REC\n");
/* Validate sgid/dgid counts against SA client limit */
p_mpr_req = (osmv_multipath_req_t *) p_query_req->p_query_input;
if (p_mpr_req->sgid_count + p_mpr_req->dgid_count >
IB_MULTIPATH_MAX_GIDS) {
OSM_LOG(p_log, OSM_LOG_ERROR, "DBG:001 MULTIPATH_REC "
"SGID count %d DGID count %d max count %d\n",
p_mpr_req->sgid_count, p_mpr_req->dgid_count,
IB_MULTIPATH_MAX_GIDS);
CL_ASSERT(0);
return IB_ERROR;
}
memset(&multipath_rec, 0, sizeof(ib_multipath_rec_t));
sa_mad_data.method = IB_MAD_METHOD_GETMULTI;
sa_mad_data.attr_id = IB_MAD_ATTR_MULTIPATH_RECORD;
sa_mad_data.attr_offset =
ib_get_attr_offset(sizeof(ib_multipath_rec_t));
sa_mad_data.p_attr = &multipath_rec;
sa_mad_data.comp_mask = p_mpr_req->comp_mask;
multipath_rec.num_path = p_mpr_req->num_path;
if (p_mpr_req->reversible)
multipath_rec.num_path |= 0x80;
else
multipath_rec.num_path &= ~0x80;
multipath_rec.pkey = p_mpr_req->pkey;
ib_multipath_rec_set_sl(&multipath_rec, p_mpr_req->sl);
ib_multipath_rec_set_qos_class(&multipath_rec, 0);
multipath_rec.independence = p_mpr_req->independence;
multipath_rec.sgid_count = p_mpr_req->sgid_count;
multipath_rec.dgid_count = p_mpr_req->dgid_count;
j = 0;
for (i = 0; i < p_mpr_req->sgid_count; i++, j++)
multipath_rec.gids[j] = p_mpr_req->gids[j];
for (i = 0; i < p_mpr_req->dgid_count; i++, j++)
multipath_rec.gids[j] = p_mpr_req->gids[j];
break;
#endif
default:
OSM_LOG(p_log, OSM_LOG_ERROR, "DBG:001 UNKNOWN\n");
CL_ASSERT(0);
return IB_ERROR;
}
status = __osmv_send_sa_req(h_bind, &sa_mad_data, p_query_req);
OSM_LOG_EXIT(p_log);
return status;
}
static ib_api_status_t mpr_rcv_get_path_parms(IN osm_sa_t * sa,
IN const ib_multipath_rec_t *
p_mpr,
IN const osm_alias_guid_t * p_src_alias_guid,
IN const osm_alias_guid_t * p_dest_alias_guid,
IN const uint16_t dest_lid_ho,
IN const ib_net64_t comp_mask,
OUT osm_path_parms_t * p_parms)
{
const osm_node_t *p_node;
const osm_physp_t *p_physp, *p_physp0;
const osm_physp_t *p_src_physp;
const osm_physp_t *p_dest_physp;
const osm_prtn_t *p_prtn = NULL;
const ib_port_info_t *p_pi, *p_pi0;
ib_slvl_table_t *p_slvl_tbl;
ib_api_status_t status = IB_SUCCESS;
uint8_t mtu;
uint8_t rate;
uint8_t pkt_life;
uint8_t required_mtu;
uint8_t required_rate;
ib_net16_t required_pkey;
uint8_t required_sl;
uint8_t required_pkt_life;
ib_net16_t dest_lid;
int hops = 0;
int in_port_num = 0;
uint8_t i;
osm_qos_level_t *p_qos_level = NULL;
uint16_t valid_sl_mask = 0xffff;
OSM_LOG_ENTER(sa->p_log);
dest_lid = cl_hton16(dest_lid_ho);
p_dest_physp = p_dest_alias_guid->p_base_port->p_physp;
p_physp = p_src_alias_guid->p_base_port->p_physp;
p_src_physp = p_physp;
p_pi = &p_physp->port_info;
mtu = ib_port_info_get_mtu_cap(p_pi);
rate = ib_port_info_compute_rate(p_pi,
p_pi->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS);
/*
Mellanox Tavor device performance is better using 1K MTU.
If required MTU and MTU selector are such that 1K is OK
and at least one end of the path is Tavor we override the
port MTU with 1K.
*/
if (sa->p_subn->opt.enable_quirks &&
sa_multipath_rec_apply_tavor_mtu_limit(p_mpr,
p_src_alias_guid->p_base_port,
p_dest_alias_guid->p_base_port,
comp_mask))
if (mtu > IB_MTU_LEN_1024) {
mtu = IB_MTU_LEN_1024;
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Optimized Path MTU to 1K for Mellanox Tavor device\n");
}
/*
Walk the subnet object from source to destination,
tracking the most restrictive rate and mtu values along the way...
If source port node is a switch, then p_physp should
point to the port that routes the destination lid
*/
p_node = osm_physp_get_node_ptr(p_physp);
if (p_node->sw) {
/*
* Source node is a switch.
* Make sure that p_physp points to the out port of the
* switch that routes to the destination lid (dest_lid_ho)
*/
p_physp = osm_switch_get_route_by_lid(p_node->sw, dest_lid);
if (p_physp == 0) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4514: "
"Can't find routing to LID %u on switch %s "
"(GUID 0x%016"PRIx64")\n", dest_lid_ho,
p_node->print_desc,
cl_ntoh64(osm_node_get_node_guid(p_node)));
status = IB_NOT_FOUND;
goto Exit;
}
}
if (sa->p_subn->opt.qos) {
/*
* Whether this node is switch or CA, the IN port for
* the sl2vl table is 0, because this is a source node.
*/
p_slvl_tbl = osm_physp_get_slvl_tbl(p_physp, 0);
/* update valid SLs that still exist on this route */
for (i = 0; i < IB_MAX_NUM_VLS; i++) {
if (valid_sl_mask & (1 << i) &&
ib_slvl_table_get(p_slvl_tbl, i) == IB_DROP_VL)
valid_sl_mask &= ~(1 << i);
}
if (!valid_sl_mask) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"All the SLs lead to VL15 on this path\n");
status = IB_NOT_FOUND;
goto Exit;
}
}
/*
* Same as above
*/
p_node = osm_physp_get_node_ptr(p_dest_physp);
if (p_node->sw) {
/*
* if destination is switch, we want p_dest_physp to point to port 0
*/
p_dest_physp =
osm_switch_get_route_by_lid(p_node->sw, dest_lid);
if (p_dest_physp == 0) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4515: "
"Can't find routing to LID %u on switch %s "
"(GUID 0x%016"PRIx64")\n", dest_lid_ho,
p_node->print_desc,
cl_ntoh64(osm_node_get_node_guid(p_node)));
status = IB_NOT_FOUND;
goto Exit;
}
}
/*
* Now go through the path step by step
*/
while (p_physp != p_dest_physp) {
int tmp_pnum = p_physp->port_num;
p_node = osm_physp_get_node_ptr(p_physp);
p_physp = osm_physp_get_remote(p_physp);
if (p_physp == 0) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4505: "
"Can't find remote phys port of %s (GUID "
"0x%016"PRIx64") port %d "
"while routing to LID %u",
p_node->print_desc,
cl_ntoh64(osm_node_get_node_guid(p_node)),
tmp_pnum,
dest_lid_ho);
status = IB_ERROR;
goto Exit;
}
/* update number of hops traversed */
hops++;
if (hops > MAX_HOPS) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4520: "
"Path from GUID 0x%016" PRIx64 " (%s) to"
" lid %u GUID 0x%016" PRIx64 " (%s) needs"
" more than %d hops, max %d hops allowed\n",
cl_ntoh64(osm_physp_get_port_guid(p_src_physp)),
p_src_physp->p_node->print_desc, dest_lid_ho,
cl_ntoh64(osm_physp_get_port_guid
(p_dest_physp)),
p_dest_physp->p_node->print_desc, hops,
MAX_HOPS);
status = IB_NOT_FOUND;
goto Exit;
}
in_port_num = osm_physp_get_port_num(p_physp);
/*
This is point to point case (no switch in between)
*/
if (p_physp == p_dest_physp)
break;
p_node = osm_physp_get_node_ptr(p_physp);
if (!p_node->sw) {
/*
There is some sort of problem in the subnet object!
If this isn't a switch, we should have reached
the destination by now!
*/
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4503: "
"Internal error, bad path while routing "
"from %s (GUID: 0x%016"PRIx64") port %d "
"to %s (GUID: 0x%016"PRIx64") port %d; "
"ended at %s port %d\n",
p_src_alias_guid->p_base_port->p_node->print_desc,
cl_ntoh64(p_src_alias_guid->p_base_port->p_node->node_info.node_guid),
p_src_alias_guid->p_base_port->p_physp->port_num,
p_dest_alias_guid->p_base_port->p_node->print_desc,
cl_ntoh64(p_dest_alias_guid->p_base_port->p_node->node_info.node_guid),
p_dest_alias_guid->p_base_port->p_physp->port_num,
p_node->print_desc,
p_physp->port_num);
status = IB_ERROR;
goto Exit;
}
/*
Check parameters for the ingress port in this switch.
*/
p_pi = &p_physp->port_info;
if (mtu > ib_port_info_get_mtu_cap(p_pi))
mtu = ib_port_info_get_mtu_cap(p_pi);
p_physp0 = osm_node_get_physp_ptr((osm_node_t *)p_node, 0);
p_pi0 = &p_physp0->port_info;
if (ib_path_compare_rates(rate,
ib_port_info_compute_rate(p_pi,
p_pi0->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS)) > 0)
rate = ib_port_info_compute_rate(p_pi,
p_pi0->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS);
/*
Continue with the egress port on this switch.
*/
p_physp = osm_switch_get_route_by_lid(p_node->sw, dest_lid);
if (p_physp == 0) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4516: "
"Dead end path on switch "
"%s (GUID: 0x%016"PRIx64") to LID %u\n",
p_node->print_desc,
cl_ntoh64(osm_node_get_node_guid(p_node)),
dest_lid_ho);
status = IB_ERROR;
goto Exit;
}
p_pi = &p_physp->port_info;
if (mtu > ib_port_info_get_mtu_cap(p_pi))
mtu = ib_port_info_get_mtu_cap(p_pi);
p_physp0 = osm_node_get_physp_ptr((osm_node_t *)p_node, 0);
p_pi0 = &p_physp0->port_info;
if (ib_path_compare_rates(rate,
ib_port_info_compute_rate(p_pi,
p_pi0->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS)) > 0)
rate = ib_port_info_compute_rate(p_pi,
p_pi0->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS);
if (sa->p_subn->opt.qos) {
/*
* Check SL2VL table of the switch and update valid SLs
*/
p_slvl_tbl =
osm_physp_get_slvl_tbl(p_physp, in_port_num);
for (i = 0; i < IB_MAX_NUM_VLS; i++) {
if (valid_sl_mask & (1 << i) &&
ib_slvl_table_get(p_slvl_tbl,
i) == IB_DROP_VL)
valid_sl_mask &= ~(1 << i);
}
if (!valid_sl_mask) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"All the SLs lead to VL15 "
"on this path\n");
status = IB_NOT_FOUND;
goto Exit;
}
}
}
/*
p_physp now points to the destination
*/
p_pi = &p_physp->port_info;
if (mtu > ib_port_info_get_mtu_cap(p_pi))
mtu = ib_port_info_get_mtu_cap(p_pi);
if (ib_path_compare_rates(rate,
ib_port_info_compute_rate(p_pi,
p_pi->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS)) > 0)
rate = ib_port_info_compute_rate(p_pi,
p_pi->capability_mask & IB_PORT_CAP_HAS_EXT_SPEEDS);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Path min MTU = %u, min rate = %u\n", mtu, rate);
/*
* Get QoS Level object according to the MultiPath request
* and adjust MultiPath parameters according to QoS settings
*/
if (sa->p_subn->opt.qos && sa->p_subn->p_qos_policy &&
(p_qos_level =
osm_qos_policy_get_qos_level_by_mpr(sa->p_subn->p_qos_policy,
p_mpr, p_src_physp,
p_dest_physp, comp_mask))) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"MultiPathRecord request matches QoS Level '%s' (%s)\n",
p_qos_level->name,
p_qos_level->use ? p_qos_level->use : "no description");
if (p_qos_level->mtu_limit_set
&& (mtu > p_qos_level->mtu_limit))
mtu = p_qos_level->mtu_limit;
if (p_qos_level->rate_limit_set
&& (ib_path_compare_rates(rate, p_qos_level->rate_limit) > 0))
rate = p_qos_level->rate_limit;
if (p_qos_level->sl_set) {
required_sl = p_qos_level->sl;
if (!(valid_sl_mask & (1 << required_sl))) {
status = IB_NOT_FOUND;
goto Exit;
}
}
}
/*
Determine if these values meet the user criteria
*/
/* we silently ignore cases where only the MTU selector is defined */
if ((comp_mask & IB_MPR_COMPMASK_MTUSELEC) &&
(comp_mask & IB_MPR_COMPMASK_MTU)) {
required_mtu = ib_multipath_rec_mtu(p_mpr);
switch (ib_multipath_rec_mtu_sel(p_mpr)) {
case 0: /* must be greater than */
if (mtu <= required_mtu)
status = IB_NOT_FOUND;
break;
case 1: /* must be less than */
if (mtu >= required_mtu) {
/* adjust to use the highest mtu
lower then the required one */
if (required_mtu > 1)
mtu = required_mtu - 1;
else
status = IB_NOT_FOUND;
}
break;
case 2: /* exact match */
if (mtu < required_mtu)
status = IB_NOT_FOUND;
else
mtu = required_mtu;
break;
case 3: /* largest available */
/* can't be disqualified by this one */
break;
default:
/* if we're here, there's a bug in ib_multipath_rec_mtu_sel() */
CL_ASSERT(FALSE);
status = IB_ERROR;
break;
}
}
if (status != IB_SUCCESS)
goto Exit;
/* we silently ignore cases where only the Rate selector is defined */
if ((comp_mask & IB_MPR_COMPMASK_RATESELEC) &&
(comp_mask & IB_MPR_COMPMASK_RATE)) {
required_rate = ib_multipath_rec_rate(p_mpr);
switch (ib_multipath_rec_rate_sel(p_mpr)) {
case 0: /* must be greater than */
if (ib_path_compare_rates(rate, required_rate) <= 0)
status = IB_NOT_FOUND;
break;
case 1: /* must be less than */
if (ib_path_compare_rates(rate, required_rate) >= 0) {
/* adjust the rate to use the highest rate
lower then the required one */
rate = ib_path_rate_get_prev(required_rate);
if (!rate)
status = IB_NOT_FOUND;
}
break;
case 2: /* exact match */
if (ib_path_compare_rates(rate, required_rate))
status = IB_NOT_FOUND;
else
rate = required_rate;
break;
case 3: /* largest available */
/* can't be disqualified by this one */
break;
default:
/* if we're here, there's a bug in ib_multipath_rec_mtu_sel() */
CL_ASSERT(FALSE);
status = IB_ERROR;
break;
}
}
if (status != IB_SUCCESS)
goto Exit;
/* Verify the pkt_life_time */
/* According to spec definition IBA 1.2 Table 205 PacketLifeTime description,
for loopback paths, packetLifeTime shall be zero. */
if (p_src_alias_guid->p_base_port == p_dest_alias_guid->p_base_port)
pkt_life = 0; /* loopback */
else if (p_qos_level && p_qos_level->pkt_life_set)
pkt_life = p_qos_level->pkt_life;
else
pkt_life = sa->p_subn->opt.subnet_timeout;
/* we silently ignore cases where only the PktLife selector is defined */
if ((comp_mask & IB_MPR_COMPMASK_PKTLIFETIMESELEC) &&
(comp_mask & IB_MPR_COMPMASK_PKTLIFETIME)) {
required_pkt_life = ib_multipath_rec_pkt_life(p_mpr);
switch (ib_multipath_rec_pkt_life_sel(p_mpr)) {
case 0: /* must be greater than */
if (pkt_life <= required_pkt_life)
status = IB_NOT_FOUND;
break;
case 1: /* must be less than */
if (pkt_life >= required_pkt_life) {
/* adjust the lifetime to use the highest possible
lower then the required one */
if (required_pkt_life > 1)
pkt_life = required_pkt_life - 1;
else
status = IB_NOT_FOUND;
}
break;
case 2: /* exact match */
if (pkt_life < required_pkt_life)
status = IB_NOT_FOUND;
else
pkt_life = required_pkt_life;
break;
case 3: /* smallest available */
/* can't be disqualified by this one */
break;
default:
/* if we're here, there's a bug in ib_path_rec_pkt_life_sel() */
CL_ASSERT(FALSE);
status = IB_ERROR;
break;
}
}
if (status != IB_SUCCESS)
goto Exit;
/*
* set Pkey for this MultiPath record request
*/
if (comp_mask & IB_MPR_COMPMASK_RAWTRAFFIC &&
cl_ntoh32(p_mpr->hop_flow_raw) & (1 << 31))
required_pkey =
osm_physp_find_common_pkey(p_src_physp, p_dest_physp,
sa->p_subn->opt.allow_both_pkeys);
else if (comp_mask & IB_MPR_COMPMASK_PKEY) {
/*
* MPR request has a specific pkey:
* Check that source and destination share this pkey.
* If QoS level has pkeys, check that this pkey exists
* in the QoS level pkeys.
* MPR returned pkey is the requested pkey.
*/
required_pkey = p_mpr->pkey;
if (!osm_physp_share_this_pkey
(p_src_physp, p_dest_physp, required_pkey,
sa->p_subn->opt.allow_both_pkeys)) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4518: "
"Ports src 0x%016"PRIx64" (%s port %d) "
"and dst 0x%016"PRIx64" (%s port %d) "
"do not share the specified PKey 0x%04x\n",
cl_ntoh64(osm_physp_get_port_guid(p_src_physp)),
p_src_physp->p_node->print_desc,
p_src_physp->port_num,
cl_ntoh64(osm_physp_get_port_guid
(p_dest_physp)),
p_dest_physp->p_node->print_desc,
p_dest_physp->port_num,
cl_ntoh16(required_pkey));
status = IB_NOT_FOUND;
goto Exit;
}
if (p_qos_level && p_qos_level->pkey_range_len &&
!osm_qos_level_has_pkey(p_qos_level, required_pkey)) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 451C: "
"Ports src 0x%016"PRIx64" (%s port %d) "
"and dst 0x%016"PRIx64" (%s port %d) "
"do not share specified PKey (0x%04x) as "
"defined by QoS level \"%s\"\n",
cl_ntoh64(osm_physp_get_port_guid(p_src_physp)),
p_src_physp->p_node->print_desc,
p_src_physp->port_num,
cl_ntoh64(osm_physp_get_port_guid
(p_dest_physp)),
p_dest_physp->p_node->print_desc,
p_dest_physp->port_num,
cl_ntoh16(required_pkey),
p_qos_level->name);
status = IB_NOT_FOUND;
goto Exit;
}
} else if (p_qos_level && p_qos_level->pkey_range_len) {
/*
* MPR request doesn't have a specific pkey, but QoS level
* has pkeys - get shared pkey from QoS level pkeys
*/
required_pkey = osm_qos_level_get_shared_pkey(p_qos_level,
p_src_physp,
p_dest_physp,
sa->p_subn->opt.allow_both_pkeys);
if (!required_pkey) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 451D: "
"Ports src 0x%016"PRIx64" (%s port %d) "
"and dst 0x%016"PRIx64" (%s port %d) "
"do not share a PKey as defined by QoS "
"level \"%s\"\n",
cl_ntoh64(osm_physp_get_port_guid(p_src_physp)),
p_src_physp->p_node->print_desc,
p_src_physp->port_num,
cl_ntoh64(osm_physp_get_port_guid
(p_dest_physp)),
p_dest_physp->p_node->print_desc,
p_dest_physp->port_num,
p_qos_level->name);
status = IB_NOT_FOUND;
goto Exit;
}
} else {
/*
* Neither MPR request nor QoS level have pkey.
* Just get any shared pkey.
*/
required_pkey =
osm_physp_find_common_pkey(p_src_physp, p_dest_physp,
sa->p_subn->opt.allow_both_pkeys);
if (!required_pkey) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4519: "
"Ports src 0x%016"PRIx64" (%s port %d) "
"and dst 0x%016"PRIx64" (%s port %d) "
"do not have any shared PKeys\n",
cl_ntoh64(osm_physp_get_port_guid(p_src_physp)),
p_src_physp->p_node->print_desc,
p_src_physp->port_num,
cl_ntoh64(osm_physp_get_port_guid
(p_dest_physp)),
p_dest_physp->p_node->print_desc,
p_dest_physp->port_num);
status = IB_NOT_FOUND;
goto Exit;
}
}
if (required_pkey) {
p_prtn =
(osm_prtn_t *) cl_qmap_get(&sa->p_subn->prtn_pkey_tbl,
required_pkey &
cl_ntoh16((uint16_t) ~ 0x8000));
if (p_prtn ==
(osm_prtn_t *) cl_qmap_end(&sa->p_subn->prtn_pkey_tbl))
p_prtn = NULL;
}
/*
* Set MultiPathRecord SL.
*/
if (comp_mask & IB_MPR_COMPMASK_SL) {
/*
* Specific SL was requested
*/
required_sl = ib_multipath_rec_sl(p_mpr);
if (p_qos_level && p_qos_level->sl_set &&
p_qos_level->sl != required_sl) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 451E: "
"QoS constraints: required MultiPathRecord SL "
"(%u) doesn't match QoS policy \"%s\" SL (%u) "
"[%s port %d <-> %s port %d]\n", required_sl,
p_qos_level->name,
p_qos_level->sl,
p_src_alias_guid->p_base_port->p_node->print_desc,
p_src_alias_guid->p_base_port->p_physp->port_num,
p_dest_alias_guid->p_base_port->p_node->print_desc,
p_dest_alias_guid->p_base_port->p_physp->port_num);
status = IB_NOT_FOUND;
goto Exit;
}
} else if (p_qos_level && p_qos_level->sl_set) {
/*
* No specific SL was requested,
* but there is an SL in QoS level.
*/
required_sl = p_qos_level->sl;
if (required_pkey && p_prtn && p_prtn->sl != p_qos_level->sl)
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"QoS level SL (%u) overrides partition SL (%u)\n",
p_qos_level->sl, p_prtn->sl);
} else if (required_pkey) {
/*
* No specific SL in request or in QoS level - use partition SL
*/
p_prtn =
(osm_prtn_t *) cl_qmap_get(&sa->p_subn->prtn_pkey_tbl,
required_pkey &
cl_ntoh16((uint16_t) ~ 0x8000));
if (!p_prtn) {
required_sl = OSM_DEFAULT_SL;
/* this may be possible when pkey tables are created somehow in
previous runs or things are going wrong here */
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 451A: "
"No partition found for PKey 0x%04x - "
"using default SL %d "
"[%s port %d <-> %s port %d]\n",
cl_ntoh16(required_pkey), required_sl,
p_src_alias_guid->p_base_port->p_node->print_desc,
p_src_alias_guid->p_base_port->p_physp->port_num,
p_dest_alias_guid->p_base_port->p_node->print_desc,
p_dest_alias_guid->p_base_port->p_physp->port_num);
} else
required_sl = p_prtn->sl;
} else if (sa->p_subn->opt.qos) {
if (valid_sl_mask & (1 << OSM_DEFAULT_SL))
required_sl = OSM_DEFAULT_SL;
else {
for (i = 0; i < IB_MAX_NUM_VLS; i++)
if (valid_sl_mask & (1 << i))
break;
required_sl = i;
}
} else
required_sl = OSM_DEFAULT_SL;
if (sa->p_subn->opt.qos && !(valid_sl_mask & (1 << required_sl))) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 451F: "
"Selected SL (%u) leads to VL15 "
"[%s port %d <-> %s port %d]\n",
required_sl,
p_src_alias_guid->p_base_port->p_node->print_desc,
p_src_alias_guid->p_base_port->p_physp->port_num,
p_dest_alias_guid->p_base_port->p_node->print_desc,
p_dest_alias_guid->p_base_port->p_physp->port_num);
status = IB_NOT_FOUND;
goto Exit;
}
/* reset pkey when raw traffic */
if (comp_mask & IB_MPR_COMPMASK_RAWTRAFFIC &&
cl_ntoh32(p_mpr->hop_flow_raw) & (1 << 31))
required_pkey = 0;
p_parms->mtu = mtu;
p_parms->rate = rate;
p_parms->pkey = required_pkey;
p_parms->pkt_life = pkt_life;
p_parms->sl = required_sl;
p_parms->hops = hops;
OSM_LOG(sa->p_log, OSM_LOG_DEBUG, "MultiPath params:"
" mtu = %u, rate = %u, packet lifetime = %u,"
" pkey = 0x%04X, sl = %u, hops = %u\n", mtu, rate,
pkt_life, cl_ntoh16(required_pkey), required_sl, hops);
Exit:
OSM_LOG_EXIT(sa->p_log);
return status;
}
void osm_mpr_rcv_process(IN void *context, IN void *data)
{
osm_sa_t *sa = context;
osm_madw_t *p_madw = data;
const ib_multipath_rec_t *p_mpr;
ib_sa_mad_t *p_sa_mad;
osm_port_t *requester_port;
osm_alias_guid_t *pp_alias_guids[IB_MULTIPATH_MAX_GIDS];
cl_qlist_t pr_list;
ib_net16_t sa_status;
int nsrc, ndest;
uint8_t rate, mtu;
OSM_LOG_ENTER(sa->p_log);
CL_ASSERT(p_madw);
p_sa_mad = osm_madw_get_sa_mad_ptr(p_madw);
p_mpr = (ib_multipath_rec_t *) ib_sa_mad_get_payload_ptr(p_sa_mad);
CL_ASSERT(p_sa_mad->attr_id == IB_MAD_ATTR_MULTIPATH_RECORD);
if ((p_sa_mad->rmpp_flags & IB_RMPP_FLAG_ACTIVE) != IB_RMPP_FLAG_ACTIVE) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4510: "
"Invalid request since RMPP_FLAG_ACTIVE is not set\n");
osm_sa_send_error(sa, p_madw, IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
/* we only support SubnAdmGetMulti method */
if (p_sa_mad->method != IB_MAD_METHOD_GETMULTI) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4513: "
"Unsupported Method (%s)\n",
ib_get_sa_method_str(p_sa_mad->method));
osm_sa_send_error(sa, p_madw, IB_MAD_STATUS_UNSUP_METHOD_ATTR);
goto Exit;
}
/* update the requester physical port */
requester_port = osm_get_port_by_mad_addr(sa->p_log, sa->p_subn,
osm_madw_get_mad_addr_ptr
(p_madw));
if (requester_port == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4517: "
"Cannot find requester physical port\n");
goto Exit;
}
if (OSM_LOG_IS_ACTIVE_V2(sa->p_log, OSM_LOG_DEBUG)) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Requester port GUID 0x%" PRIx64 "\n",
cl_ntoh64(osm_port_get_guid(requester_port)));
osm_dump_multipath_record_v2(sa->p_log, p_mpr, FILE_ID, OSM_LOG_DEBUG);
}
/* Validatg rate if supplied */
if ((p_sa_mad->comp_mask & IB_MPR_COMPMASK_RATESELEC) &&
(p_sa_mad->comp_mask & IB_MPR_COMPMASK_RATE)) {
rate = ib_multipath_rec_rate(p_mpr);
if (!ib_rate_is_valid(rate)) {
osm_sa_send_error(sa, p_madw,
IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
}
/* Validate MTU if supplied */
if ((p_sa_mad->comp_mask & IB_MPR_COMPMASK_MTUSELEC) &&
(p_sa_mad->comp_mask & IB_MPR_COMPMASK_MTU)) {
mtu = ib_multipath_rec_mtu(p_mpr);
if (!ib_mtu_is_valid(mtu)) {
osm_sa_send_error(sa, p_madw,
IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
}
cl_qlist_init(&pr_list);
/*
Most SA functions (including this one) are read-only on the
subnet object, so we grab the lock non-exclusively.
*/
cl_plock_acquire(sa->p_lock);
sa_status = mpr_rcv_get_end_points(sa, p_madw, pp_alias_guids,
&nsrc, &ndest);
if (sa_status != IB_SA_MAD_STATUS_SUCCESS || !nsrc || !ndest) {
if (sa_status == IB_SA_MAD_STATUS_SUCCESS && (!nsrc || !ndest))
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4512: "
"mpr_rcv_get_end_points failed, # GIDs found; "
"src %d; dest %d)\n", nsrc, ndest);
cl_plock_release(sa->p_lock);
if (sa_status == IB_SA_MAD_STATUS_SUCCESS)
osm_sa_send_error(sa, p_madw,
IB_SA_MAD_STATUS_REQ_INVALID);
else
osm_sa_send_error(sa, p_madw, sa_status);
goto Exit;
}
/* APM request */
if (nsrc == 2 && ndest == 2 && (p_mpr->num_path & 0x7F) == 2)
mpr_rcv_get_apm_paths(sa, p_mpr, requester_port, pp_alias_guids,
p_sa_mad->comp_mask, &pr_list);
else
mpr_rcv_process_pairs(sa, p_mpr, requester_port, pp_alias_guids,
nsrc, ndest, p_sa_mad->comp_mask,
&pr_list);
cl_plock_release(sa->p_lock);
/* o15-0.2.7: If MultiPath is supported, then SA shall respond to a
SubnAdmGetMulti() containing a valid MultiPathRecord attribute with
a set of zero or more PathRecords satisfying the constraints
indicated in the MultiPathRecord received. The PathRecord Attribute
ID shall be used in the response.
*/
p_sa_mad->attr_id = IB_MAD_ATTR_PATH_RECORD;
osm_sa_respond(sa, p_madw, sizeof(ib_path_rec_t), &pr_list);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
static osm_sa_item_t *mpr_rcv_get_apm_port_pair_paths(IN osm_sa_t * sa,
IN const
ib_multipath_rec_t *
p_mpr,
IN const osm_alias_guid_t *
p_src_alias_guid,
IN const osm_alias_guid_t *
p_dest_alias_guid,
IN int base_offs,
IN const ib_net64_t
comp_mask,
IN cl_qlist_t * p_list)
{
osm_sa_item_t *p_pr_item = 0;
uint16_t src_lid_min_ho;
uint16_t src_lid_max_ho;
uint16_t dest_lid_min_ho;
uint16_t dest_lid_max_ho;
uint16_t src_lid_ho;
uint16_t dest_lid_ho;
unsigned iterations;
int src_lids, dest_lids;
OSM_LOG_ENTER(sa->p_log);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG, "Src port 0x%016" PRIx64 ", "
"Dst port 0x%016" PRIx64 ", base offs %d\n",
cl_ntoh64(p_src_alias_guid->alias_guid),
cl_ntoh64(p_dest_alias_guid->alias_guid),
base_offs);
osm_port_get_lid_range_ho(p_src_alias_guid->p_base_port,
&src_lid_min_ho, &src_lid_max_ho);
osm_port_get_lid_range_ho(p_dest_alias_guid->p_base_port,
&dest_lid_min_ho, &dest_lid_max_ho);
src_lid_ho = src_lid_min_ho;
dest_lid_ho = dest_lid_min_ho;
src_lids = src_lid_max_ho - src_lid_min_ho + 1;
dest_lids = dest_lid_max_ho - dest_lid_min_ho + 1;
src_lid_ho += base_offs % src_lids;
dest_lid_ho += base_offs % dest_lids;
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Src LIDs [%u-%u] hashed %u, "
"Dest LIDs [%u-%u] hashed %u\n",
src_lid_min_ho, src_lid_max_ho, src_lid_ho,
dest_lid_min_ho, dest_lid_max_ho, dest_lid_ho);
iterations = min(src_lids, dest_lids);
while (iterations--) {
/*
These paths are "fully redundant"
*/
p_pr_item = mpr_rcv_get_lid_pair_path(sa, p_mpr,
p_src_alias_guid,
p_dest_alias_guid,
src_lid_ho, dest_lid_ho,
comp_mask, 0);
if (p_pr_item) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Found matching path from Src LID %u to Dest LID %u with %d hops\n",
src_lid_ho, dest_lid_ho, p_pr_item->resp.mpr_rec.hops);
break;
}
if (++src_lid_ho > src_lid_max_ho)
src_lid_ho = src_lid_min_ho;
if (++dest_lid_ho > dest_lid_max_ho)
dest_lid_ho = dest_lid_min_ho;
}
OSM_LOG_EXIT(sa->p_log);
return p_pr_item;
}
static void mpr_rcv_get_apm_paths(IN osm_sa_t * sa,
IN const ib_multipath_rec_t * p_mpr,
IN const osm_port_t * p_req_port,
IN osm_alias_guid_t ** _pp_alias_guids,
IN const ib_net64_t comp_mask,
IN cl_qlist_t * p_list)
{
osm_alias_guid_t *pp_alias_guids[4];
osm_sa_item_t *matrix[2][2];
int base_offs, src_lid_ho, dest_lid_ho;
int sumA, sumB, minA, minB;
OSM_LOG_ENTER(sa->p_log);
/*
* We want to:
* 1. use different lid offsets (from base) for the resultant paths
* to increase the probability of redundant paths or in case
* of Clos - to ensure it (different offset => different spine!)
* 2. keep consistent paths no matter of direction and order of ports
* 3. distibute the lid offsets to balance the load
* So, we sort the ports (within the srcs, and within the dests),
* hash the lids of S0, D0 (after the sort), and call mpr_rcv_get_apm_port_pair_paths
* with base_lid for S0, D0 and base_lid + 1 for S1, D1. This way we will get
* always the same offsets - order independent, and make sure different spines are used.
* Note that the diagonals on a Clos have the same number of hops, so it doesn't
* really matter which diagonal we use.
*/
if (_pp_alias_guids[0]->p_base_port->guid <
_pp_alias_guids[1]->p_base_port->guid) {
pp_alias_guids[0] = _pp_alias_guids[0];
pp_alias_guids[1] = _pp_alias_guids[1];
} else {
pp_alias_guids[0] = _pp_alias_guids[1];
pp_alias_guids[1] = _pp_alias_guids[0];
}
if (_pp_alias_guids[2]->p_base_port->guid <
_pp_alias_guids[3]->p_base_port->guid) {
pp_alias_guids[2] = _pp_alias_guids[2];
pp_alias_guids[3] = _pp_alias_guids[3];
} else {
pp_alias_guids[2] = _pp_alias_guids[3];
pp_alias_guids[3] = _pp_alias_guids[2];
}
src_lid_ho = osm_port_get_base_lid(pp_alias_guids[0]->p_base_port);
dest_lid_ho = osm_port_get_base_lid(pp_alias_guids[2]->p_base_port);
base_offs = src_lid_ho < dest_lid_ho ?
hash_lids(src_lid_ho, dest_lid_ho, sa->p_subn->opt.lmc) :
hash_lids(dest_lid_ho, src_lid_ho, sa->p_subn->opt.lmc);
matrix[0][0] =
mpr_rcv_get_apm_port_pair_paths(sa, p_mpr, pp_alias_guids[0],
pp_alias_guids[2], base_offs,
comp_mask, p_list);
matrix[0][1] =
mpr_rcv_get_apm_port_pair_paths(sa, p_mpr, pp_alias_guids[0],
pp_alias_guids[3], base_offs,
comp_mask, p_list);
matrix[1][0] =
mpr_rcv_get_apm_port_pair_paths(sa, p_mpr, pp_alias_guids[1],
pp_alias_guids[2], base_offs + 1,
comp_mask, p_list);
matrix[1][1] =
mpr_rcv_get_apm_port_pair_paths(sa, p_mpr, pp_alias_guids[1],
pp_alias_guids[3], base_offs + 1,
comp_mask, p_list);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG, "APM matrix:\n"
"\t{0,0} 0x%X->0x%X (%d)\t| {0,1} 0x%X->0x%X (%d)\n"
"\t{1,0} 0x%X->0x%X (%d)\t| {1,1} 0x%X->0x%X (%d)\n",
matrix[0][0]->resp.mpr_rec.path_rec.slid,
matrix[0][0]->resp.mpr_rec.path_rec.dlid,
matrix[0][0]->resp.mpr_rec.hops,
matrix[0][1]->resp.mpr_rec.path_rec.slid,
matrix[0][1]->resp.mpr_rec.path_rec.dlid,
matrix[0][1]->resp.mpr_rec.hops,
matrix[1][0]->resp.mpr_rec.path_rec.slid,
matrix[1][0]->resp.mpr_rec.path_rec.dlid,
matrix[1][0]->resp.mpr_rec.hops,
matrix[1][1]->resp.mpr_rec.path_rec.slid,
matrix[1][1]->resp.mpr_rec.path_rec.dlid,
matrix[1][1]->resp.mpr_rec.hops);
/* check diagonal A {(0,0), (1,1)} */
sumA = matrix[0][0]->resp.mpr_rec.hops + matrix[1][1]->resp.mpr_rec.hops;
minA = min(matrix[0][0]->resp.mpr_rec.hops, matrix[1][1]->resp.mpr_rec.hops);
/* check diagonal B {(0,1), (1,0)} */
sumB = matrix[0][1]->resp.mpr_rec.hops + matrix[1][0]->resp.mpr_rec.hops;
minB = min(matrix[0][1]->resp.mpr_rec.hops, matrix[1][0]->resp.mpr_rec.hops);
/* and the winner is... */
if (minA <= minB || (minA == minB && sumA < sumB)) {
/* Diag A */
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Diag {0,0} & {1,1} is the best:\n"
"\t{0,0} 0x%X->0x%X (%d)\t & {1,1} 0x%X->0x%X (%d)\n",
matrix[0][0]->resp.mpr_rec.path_rec.slid,
matrix[0][0]->resp.mpr_rec.path_rec.dlid,
matrix[0][0]->resp.mpr_rec.hops,
matrix[1][1]->resp.mpr_rec.path_rec.slid,
matrix[1][1]->resp.mpr_rec.path_rec.dlid,
matrix[1][1]->resp.mpr_rec.hops);
cl_qlist_insert_tail(p_list, &matrix[0][0]->list_item);
cl_qlist_insert_tail(p_list, &matrix[1][1]->list_item);
free(matrix[0][1]);
free(matrix[1][0]);
} else {
/* Diag B */
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Diag {0,1} & {1,0} is the best:\n"
"\t{0,1} 0x%X->0x%X (%d)\t & {1,0} 0x%X->0x%X (%d)\n",
matrix[0][1]->resp.mpr_rec.path_rec.slid,
matrix[0][1]->resp.mpr_rec.path_rec.dlid,
matrix[0][1]->resp.mpr_rec.hops,
matrix[1][0]->resp.mpr_rec.path_rec.slid,
matrix[1][0]->resp.mpr_rec.path_rec.dlid,
matrix[1][0]->resp.mpr_rec.hops);
cl_qlist_insert_tail(p_list, &matrix[0][1]->list_item);
cl_qlist_insert_tail(p_list, &matrix[1][0]->list_item);
free(matrix[0][0]);
free(matrix[1][1]);
}
OSM_LOG_EXIT(sa->p_log);
}
void osm_lftr_rcv_process(IN void *ctx, IN void *data)
{
osm_sa_t *sa = ctx;
osm_madw_t *p_madw = data;
const ib_sa_mad_t *p_rcvd_mad;
const ib_lft_record_t *p_rcvd_rec;
cl_qlist_t rec_list;
osm_lftr_search_ctxt_t context;
osm_physp_t *p_req_physp;
CL_ASSERT(sa);
OSM_LOG_ENTER(sa->p_log);
CL_ASSERT(p_madw);
p_rcvd_mad = osm_madw_get_sa_mad_ptr(p_madw);
p_rcvd_rec = (ib_lft_record_t *) ib_sa_mad_get_payload_ptr(p_rcvd_mad);
CL_ASSERT(p_rcvd_mad->attr_id == IB_MAD_ATTR_LFT_RECORD);
/* we only support SubnAdmGet and SubnAdmGetTable methods */
if (p_rcvd_mad->method != IB_MAD_METHOD_GET &&
p_rcvd_mad->method != IB_MAD_METHOD_GETTABLE) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4408: "
"Unsupported Method (%s)\n",
ib_get_sa_method_str(p_rcvd_mad->method));
osm_sa_send_error(sa, p_madw, IB_MAD_STATUS_UNSUP_METHOD_ATTR);
goto Exit;
}
/* update the requester physical port */
p_req_physp = osm_get_physp_by_mad_addr(sa->p_log, sa->p_subn,
osm_madw_get_mad_addr_ptr
(p_madw));
if (p_req_physp == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4407: "
"Cannot find requester physical port\n");
goto Exit;
}
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Requester port GUID 0x%" PRIx64 "\n",
cl_ntoh64(osm_physp_get_port_guid(p_req_physp)));
cl_qlist_init(&rec_list);
context.p_rcvd_rec = p_rcvd_rec;
context.p_list = &rec_list;
context.comp_mask = p_rcvd_mad->comp_mask;
context.sa = sa;
context.p_req_physp = p_req_physp;
cl_plock_acquire(sa->p_lock);
/* Go over all switches */
cl_qmap_apply_func(&sa->p_subn->sw_guid_tbl, lftr_rcv_by_comp_mask,
&context);
cl_plock_release(sa->p_lock);
osm_sa_respond(sa, p_madw, sizeof(ib_lft_record_t), &rec_list);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
static void lftr_rcv_by_comp_mask(IN cl_map_item_t * p_map_item, IN void *cxt)
{
const osm_lftr_search_ctxt_t *p_ctxt = cxt;
const osm_switch_t *p_sw = (osm_switch_t *) p_map_item;
const ib_lft_record_t *const p_rcvd_rec = p_ctxt->p_rcvd_rec;
osm_sa_t *sa = p_ctxt->sa;
ib_net64_t const comp_mask = p_ctxt->comp_mask;
const osm_physp_t *const p_req_physp = p_ctxt->p_req_physp;
osm_port_t *p_port;
uint16_t min_lid_ho, max_lid_ho;
uint16_t min_block, max_block, block;
const osm_physp_t *p_physp;
/* In switches, the port guid is the node guid. */
p_port = osm_get_port_by_guid(sa->p_subn,
p_sw->p_node->node_info.port_guid);
if (!p_port) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4405: "
"Failed to find Port by Node Guid:0x%016" PRIx64
"\n", cl_ntoh64(p_sw->p_node->node_info.node_guid));
return;
}
/* check that the requester physp and the current physp are under
the same partition. */
p_physp = p_port->p_physp;
if (!p_physp) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4406: "
"Failed to find default physical Port by Node Guid:0x%016"
PRIx64 "\n",
cl_ntoh64(p_sw->p_node->node_info.node_guid));
return;
}
if (!osm_physp_share_pkey(sa->p_log, p_req_physp,
p_physp, sa->p_subn->opt.allow_both_pkeys))
return;
/* get the port 0 of the switch */
osm_port_get_lid_range_ho(p_port, &min_lid_ho, &max_lid_ho);
/* compare the lids - if required */
if (comp_mask & IB_LFTR_COMPMASK_LID) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Comparing lid:%u to port lid range: %u .. %u\n",
cl_ntoh16(p_rcvd_rec->lid), min_lid_ho, max_lid_ho);
/* ok we are ready for range check */
if (min_lid_ho > cl_ntoh16(p_rcvd_rec->lid) ||
max_lid_ho < cl_ntoh16(p_rcvd_rec->lid))
return;
}
/* now we need to decide which blocks to output */
max_block = osm_switch_get_max_block_id_in_use(p_sw);
if (comp_mask & IB_LFTR_COMPMASK_BLOCK) {
min_block = cl_ntoh16(p_rcvd_rec->block_num);
if (min_block > max_block)
return;
max_block = min_block;
} else /* use as many blocks as "in use" */
min_block = 0;
/* so we can add these blocks one by one ... */
for (block = min_block; block <= max_block; block++)
lftr_rcv_new_lftr(sa, p_sw, p_ctxt->p_list,
osm_port_get_base_lid(p_port), block);
}
/**********************************************************************
* Make sure that the lid_port_tbl of the subnet has only the ports
* that are recognized, and in the correct lid place. There could be
* errors if we wanted to assign a certain port with lid X, but that
* request didn't reach the port. In this case port_lid_tbl will have
* the port under lid X, though the port isn't updated with this lid.
* We will run a new heavy sweep (since there were errors in the
* initialization), but here we'll clean the database from incorrect
* information.
**********************************************************************/
static void state_mgr_check_tbl_consistency(IN osm_sm_t * sm)
{
cl_qmap_t *p_port_guid_tbl;
osm_port_t *p_port;
osm_port_t *p_next_port;
cl_ptr_vector_t *p_port_lid_tbl;
size_t max_lid, ref_size, curr_size, lid;
osm_port_t *p_port_ref, *p_port_stored;
cl_ptr_vector_t ref_port_lid_tbl;
uint16_t min_lid_ho;
uint16_t max_lid_ho;
uint16_t lid_ho;
OSM_LOG_ENTER(sm->p_log);
cl_ptr_vector_construct(&ref_port_lid_tbl);
cl_ptr_vector_init(&ref_port_lid_tbl,
cl_ptr_vector_get_size(&sm->p_subn->port_lid_tbl),
OSM_SUBNET_VECTOR_GROW_SIZE);
p_port_guid_tbl = &sm->p_subn->port_guid_tbl;
/* Let's go over all the ports according to port_guid_tbl,
* and add the port to a reference port_lid_tbl. */
p_next_port = (osm_port_t *) cl_qmap_head(p_port_guid_tbl);
while (p_next_port != (osm_port_t *) cl_qmap_end(p_port_guid_tbl)) {
p_port = p_next_port;
p_next_port =
(osm_port_t *) cl_qmap_next(&p_next_port->map_item);
osm_port_get_lid_range_ho(p_port, &min_lid_ho, &max_lid_ho);
for (lid_ho = min_lid_ho; lid_ho <= max_lid_ho; lid_ho++)
cl_ptr_vector_set(&ref_port_lid_tbl, lid_ho, p_port);
}
p_port_lid_tbl = &sm->p_subn->port_lid_tbl;
ref_size = cl_ptr_vector_get_size(&ref_port_lid_tbl);
curr_size = cl_ptr_vector_get_size(p_port_lid_tbl);
/* They should be the same, but compare it anyway */
max_lid = (ref_size > curr_size) ? ref_size : curr_size;
for (lid = 1; lid < max_lid; lid++) {
p_port_ref = NULL;
p_port_stored = NULL;
cl_ptr_vector_at(p_port_lid_tbl, lid, (void *)&p_port_stored);
cl_ptr_vector_at(&ref_port_lid_tbl, lid, (void *)&p_port_ref);
if (p_port_stored == p_port_ref)
/* This is the "good" case - both entries are the
* same for this lid. Nothing to do. */
continue;
if (p_port_ref == NULL)
/* There is an object in the subnet database for this
* lid, but no such object exists in the reference
* port_list_tbl. This can occur if we wanted to assign
* a certain port with some lid (different than the one
* pre-assigned to it), and the port didn't get the
* PortInfo Set request. Due to this, the port is
* updated with its original lid in our database, but
* with the new lid we wanted to give it in our
* port_lid_tbl. */
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3322: "
"lid %zu is wrongly assigned to port 0x%016"
PRIx64 " (\'%s\' port %u) in port_lid_tbl\n",
lid,
cl_ntoh64(osm_port_get_guid(p_port_stored)),
p_port_stored->p_node->print_desc,
p_port_stored->p_physp->port_num);
else if (p_port_stored == NULL)
/* There is an object in the new database, but no
* object in our subnet database. This is the matching
* case of the prior check - the port still has its
* original lid. */
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3323: "
"port 0x%016" PRIx64 " (\'%s\' port %u)"
" exists in new port_lid_tbl under lid %zu,"
" but missing in subnet port_lid_tbl db\n",
cl_ntoh64(osm_port_get_guid(p_port_ref)),
p_port_ref->p_node->print_desc,
p_port_ref->p_physp->port_num, lid);
else
/* if we reached here then p_port_stored != p_port_ref.
* We were trying to set a lid to p_port_stored, but
* it didn't reach it, and p_port_ref also didn't get
* the lid update. */
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3324: "
"lid %zu has port 0x%016" PRIx64
" (\'%s\' port %u) in new port_lid_tbl db, "
"and port 0x%016" PRIx64 " (\'%s\' port %u)"
" in subnet port_lid_tbl db\n", lid,
cl_ntoh64(osm_port_get_guid(p_port_ref)),
p_port_ref->p_node->print_desc,
p_port_ref->p_physp->port_num,
cl_ntoh64(osm_port_get_guid(p_port_stored)),
p_port_ref->p_node->print_desc,
p_port_ref->p_physp->port_num);
/* In any of these cases we want to set NULL in the
* port_lid_tbl, since this entry is invalid. Also, make sure
* we'll do another heavy sweep. */
cl_ptr_vector_set(p_port_lid_tbl, lid, NULL);
sm->p_subn->subnet_initialization_error = TRUE;
}
cl_ptr_vector_destroy(&ref_port_lid_tbl);
OSM_LOG_EXIT(sm->p_log);
}
/**********************************************************************
o13-14.1.1: Except for Set(InformInfo) requests with Inform-
Info:LIDRangeBegin=0xFFFF, managers that support event forwarding
shall, upon receiving a Set(InformInfo), verify that the requester
originating the Set(InformInfo) and a Trap() source identified by Inform-
can access each other - can use path record to verify that.
**********************************************************************/
static boolean_t
__validate_ports_access_rights(IN osm_sa_t * sa,
IN osm_infr_t * p_infr_rec)
{
boolean_t valid = TRUE;
osm_physp_t *p_requester_physp;
osm_port_t *p_port;
osm_physp_t *p_physp;
ib_net64_t portguid;
ib_net16_t lid_range_begin;
ib_net16_t lid_range_end;
ib_net16_t lid;
const cl_ptr_vector_t *p_tbl;
ib_gid_t zero_gid;
OSM_LOG_ENTER(sa->p_log);
/* get the requester physp from the request address */
p_requester_physp = osm_get_physp_by_mad_addr(sa->p_log, sa->p_subn,
&p_infr_rec->report_addr);
memset(&zero_gid, 0, sizeof(zero_gid));
if (memcmp(&(p_infr_rec->inform_record.inform_info.gid),
&zero_gid, sizeof(ib_gid_t))) {
/* a gid is defined */
portguid =
p_infr_rec->inform_record.inform_info.gid.unicast.
interface_id;
p_port = osm_get_port_by_guid(sa->p_subn, portguid);
if (p_port == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4301: "
"Invalid port guid: 0x%016" PRIx64 "\n",
cl_ntoh64(portguid));
valid = FALSE;
goto Exit;
}
/* get the destination InformInfo physical port */
p_physp = p_port->p_physp;
/* make sure that the requester and destination port can access each other
according to the current partitioning. */
if (!osm_physp_share_pkey
(sa->p_log, p_physp, p_requester_physp)) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"port and requester don't share pkey\n");
valid = FALSE;
goto Exit;
}
} else {
/* gid is zero - check if LID range is defined */
lid_range_begin =
cl_ntoh16(p_infr_rec->inform_record.inform_info.
lid_range_begin);
/* if lid is 0xFFFF - meaning all endports managed by the manager */
if (lid_range_begin == 0xFFFF)
goto Exit;
lid_range_end =
cl_ntoh16(p_infr_rec->inform_record.inform_info.
lid_range_end);
/* lid_range_end is set to zero if no range desired. In this case -
just make it equal to the lid_range_begin. */
if (lid_range_end == 0)
lid_range_end = lid_range_begin;
/* go over all defined lids within the range and make sure that the
requester port can access them according to current partitioning. */
for (lid = lid_range_begin; lid <= lid_range_end; lid++) {
p_tbl = &sa->p_subn->port_lid_tbl;
if (cl_ptr_vector_get_size(p_tbl) > lid)
p_port = cl_ptr_vector_get(p_tbl, lid);
else {
/* lid requested is out of range */
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4302: "
"Given LID (%u) is out of range:%u\n",
lid, cl_ptr_vector_get_size(p_tbl));
valid = FALSE;
goto Exit;
}
if (p_port == NULL)
continue;
p_physp = p_port->p_physp;
/* make sure that the requester and destination port can access
each other according to the current partitioning. */
if (!osm_physp_share_pkey
(sa->p_log, p_physp, p_requester_physp)) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"port and requester don't share pkey\n");
valid = FALSE;
goto Exit;
}
}
}
Exit:
OSM_LOG_EXIT(sa->p_log);
return valid;
}
static osm_sa_item_t *mpr_rcv_get_lid_pair_path(IN osm_sa_t * sa,
IN const ib_multipath_rec_t *
p_mpr,
IN const osm_alias_guid_t *
p_src_alias_guid,
IN const osm_alias_guid_t *
p_dest_alias_guid,
IN const uint16_t src_lid_ho,
IN const uint16_t dest_lid_ho,
IN const ib_net64_t comp_mask,
IN const uint8_t preference)
{
osm_path_parms_t path_parms;
osm_path_parms_t rev_path_parms;
osm_sa_item_t *p_pr_item;
ib_api_status_t status, rev_path_status;
OSM_LOG_ENTER(sa->p_log);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG, "Src LID %u, Dest LID %u\n",
src_lid_ho, dest_lid_ho);
p_pr_item = malloc(SA_MPR_RESP_SIZE);
if (p_pr_item == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4501: "
"Unable to allocate path record\n");
goto Exit;
}
memset(p_pr_item, 0, SA_MPR_RESP_SIZE);
status = mpr_rcv_get_path_parms(sa, p_mpr, p_src_alias_guid,
p_dest_alias_guid, dest_lid_ho,
comp_mask, &path_parms);
if (status != IB_SUCCESS) {
free(p_pr_item);
p_pr_item = NULL;
goto Exit;
}
/* now try the reversible path */
rev_path_status = mpr_rcv_get_path_parms(sa, p_mpr, p_dest_alias_guid,
p_src_alias_guid, src_lid_ho,
comp_mask, &rev_path_parms);
path_parms.reversible = (rev_path_status == IB_SUCCESS);
/* did we get a Reversible Path compmask ? */
/*
NOTE that if the reversible component = 0, it is a don't care
rather then requiring non-reversible paths ...
see Vol1 Ver1.2 p900 l16
*/
if (comp_mask & IB_MPR_COMPMASK_REVERSIBLE) {
if ((!path_parms.reversible && (p_mpr->num_path & 0x80))) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Requested reversible path but failed to get one\n");
free(p_pr_item);
p_pr_item = NULL;
goto Exit;
}
}
p_pr_item->resp.mpr_rec.p_src_port = p_src_alias_guid->p_base_port;
p_pr_item->resp.mpr_rec.p_dest_port = p_dest_alias_guid->p_base_port;
p_pr_item->resp.mpr_rec.hops = path_parms.hops;
mpr_rcv_build_pr(sa, p_src_alias_guid, p_dest_alias_guid, src_lid_ho,
dest_lid_ho, preference, &path_parms,
&p_pr_item->resp.mpr_rec.path_rec);
Exit:
OSM_LOG_EXIT(sa->p_log);
return p_pr_item;
}
static int link_mgr_process_node(osm_sm_t * sm, IN osm_node_t * p_node,
IN const uint8_t link_state)
{
osm_physp_t *p_physp, *p_physp_remote;
uint32_t i, num_physp;
int ret = 0;
uint8_t current_state;
OSM_LOG_ENTER(sm->p_log);
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Node 0x%" PRIx64 " going to %s\n",
cl_ntoh64(osm_node_get_node_guid(p_node)),
ib_get_port_state_str(link_state));
/*
Set the PortInfo for every Physical Port associated
with this Port. Start iterating with port 1, since the linkstate
is not applicable to the management port on switches.
*/
num_physp = osm_node_get_num_physp(p_node);
for (i = 0; i < num_physp; i++) {
/*
Don't bother doing anything if this Physical Port is not valid.
or if the state of the port is already better then the
specified state.
*/
p_physp = osm_node_get_physp_ptr(p_node, (uint8_t) i);
if (!p_physp)
continue;
current_state = osm_physp_get_port_state(p_physp);
if (current_state == IB_LINK_DOWN)
continue;
/*
Set PortState to DOWN in case Remote Physical Port is
unreachable. We have to check this for all ports, except
port zero.
*/
p_physp_remote = osm_physp_get_remote(p_physp);
if ((i != 0) && (!p_physp_remote ||
!osm_physp_is_valid(p_physp_remote))) {
if (current_state != IB_LINK_INIT)
link_mgr_set_physp_pi(sm, p_physp, IB_LINK_DOWN);
continue;
}
/*
Normally we only send state update if state is lower
then required state. However, we need to send update if
no state change required.
*/
if (link_state != IB_LINK_NO_CHANGE &&
link_state <= current_state)
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Physical port %u already %s. Skipping\n",
p_physp->port_num,
ib_get_port_state_str(current_state));
else if (link_mgr_set_physp_pi(sm, p_physp, link_state))
ret = -1;
}
OSM_LOG_EXIT(sm->p_log);
return ret;
}
static uint32_t mpr_rcv_get_port_pair_paths(IN osm_sa_t * sa,
IN const ib_multipath_rec_t * p_mpr,
IN const osm_port_t * p_req_port,
IN const osm_alias_guid_t * p_src_alias_guid,
IN const osm_alias_guid_t * p_dest_alias_guid,
IN const uint32_t rem_paths,
IN const ib_net64_t comp_mask,
IN cl_qlist_t * p_list)
{
osm_sa_item_t *p_pr_item;
uint16_t src_lid_min_ho;
uint16_t src_lid_max_ho;
uint16_t dest_lid_min_ho;
uint16_t dest_lid_max_ho;
uint16_t src_lid_ho;
uint16_t dest_lid_ho;
uint32_t path_num = 0;
uint8_t preference;
unsigned src_offset, dest_offset;
OSM_LOG_ENTER(sa->p_log);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Src port 0x%016" PRIx64 ", Dst port 0x%016" PRIx64 "\n",
cl_ntoh64(p_src_alias_guid->alias_guid),
cl_ntoh64(p_dest_alias_guid->alias_guid));
/* Check that the req_port, src_port and dest_port all share a
pkey. The check is done on the default physical port of the ports. */
if (osm_port_share_pkey(sa->p_log, p_req_port,
p_src_alias_guid->p_base_port,
sa->p_subn->opt.allow_both_pkeys) == FALSE
|| osm_port_share_pkey(sa->p_log, p_req_port,
p_dest_alias_guid->p_base_port,
sa->p_subn->opt.allow_both_pkeys) == FALSE
|| osm_port_share_pkey(sa->p_log, p_src_alias_guid->p_base_port,
p_dest_alias_guid->p_base_port,
sa->p_subn->opt.allow_both_pkeys) == FALSE)
/* One of the pairs doesn't share a pkey so the path is disqualified. */
goto Exit;
/*
We shouldn't be here if the paths are disqualified in some way...
Thus, we assume every possible connection is valid.
We desire to return high-quality paths first.
In OpenSM, higher quality mean least overlap with other paths.
This is acheived in practice by returning paths with
different LID value on each end, which means these
paths are more redundant that paths with the same LID repeated
on one side. For example, in OpenSM the paths between two
endpoints with LMC = 1 might be as follows:
Port A, LID 1 <-> Port B, LID 3
Port A, LID 1 <-> Port B, LID 4
Port A, LID 2 <-> Port B, LID 3
Port A, LID 2 <-> Port B, LID 4
The OpenSM unicast routing algorithms attempt to disperse each path
to as varied a physical path as is reasonable. 1<->3 and 1<->4 have
more physical overlap (hence less redundancy) than 1<->3 and 2<->4.
OpenSM ranks paths in three preference groups:
Preference Value Description
---------------- -------------------------------------------
0 Redundant in both directions with other
pref value = 0 paths
1 Redundant in one direction with other
pref value = 0 and pref value = 1 paths
2 Not redundant in either direction with
other paths
3-FF Unused
SA clients don't need to know these details, only that the lower
preference paths are preferred, as stated in the spec. The paths
may not actually be physically redundant depending on the topology
of the subnet, but the point of LMC > 0 is to offer redundancy,
so I assume the subnet is physically appropriate for the specified
LMC value. A more advanced implementation could inspect for physical
redundancy, but I'm not going to bother with that now.
*/
osm_port_get_lid_range_ho(p_src_alias_guid->p_base_port,
&src_lid_min_ho, &src_lid_max_ho);
osm_port_get_lid_range_ho(p_dest_alias_guid->p_base_port,
&dest_lid_min_ho, &dest_lid_max_ho);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG, "Src LID [%u-%u], Dest LID [%u-%u]\n",
src_lid_min_ho, src_lid_max_ho,
dest_lid_min_ho, dest_lid_max_ho);
src_lid_ho = src_lid_min_ho;
dest_lid_ho = dest_lid_min_ho;
/*
Preferred paths come first in OpenSM
*/
preference = 0;
while (path_num < rem_paths) {
/*
These paths are "fully redundant"
*/
p_pr_item = mpr_rcv_get_lid_pair_path(sa, p_mpr,
p_src_alias_guid,
p_dest_alias_guid,
src_lid_ho, dest_lid_ho,
comp_mask, preference);
if (p_pr_item) {
cl_qlist_insert_tail(p_list, &p_pr_item->list_item);
++path_num;
}
if (++src_lid_ho > src_lid_max_ho)
break;
if (++dest_lid_ho > dest_lid_max_ho)
break;
}
/*
Check if we've accumulated all the paths that the user cares to see
*/
if (path_num == rem_paths)
goto Exit;
/*
Don't bother reporting preference 1 paths for now.
It's more trouble than it's worth and can only occur
if ports have different LMC values, which isn't supported
by OpenSM right now anyway.
*/
preference = 2;
src_lid_ho = src_lid_min_ho;
dest_lid_ho = dest_lid_min_ho;
src_offset = 0;
dest_offset = 0;
/*
Iterate over the remaining paths
*/
while (path_num < rem_paths) {
dest_offset++;
dest_lid_ho++;
if (dest_lid_ho > dest_lid_max_ho) {
src_offset++;
src_lid_ho++;
if (src_lid_ho > src_lid_max_ho)
break; /* done */
dest_offset = 0;
dest_lid_ho = dest_lid_min_ho;
}
/*
These paths are "fully non-redundant" with paths already
identified above and consequently not of much value.
Don't return paths we already identified above, as indicated
by the offset values being equal.
*/
if (src_offset == dest_offset)
continue; /* already reported */
p_pr_item = mpr_rcv_get_lid_pair_path(sa, p_mpr,
p_src_alias_guid,
p_dest_alias_guid,
src_lid_ho, dest_lid_ho,
comp_mask, preference);
if (p_pr_item) {
cl_qlist_insert_tail(p_list, &p_pr_item->list_item);
++path_num;
}
}
Exit:
OSM_LOG_EXIT(sa->p_log);
return path_num;
}
/* Send a MAD out on the GSI interface */
static ib_api_status_t
__osmv_send_sa_req(IN osmv_sa_bind_info_t * p_bind,
IN const osmv_sa_mad_data_t * const p_sa_mad_data,
IN const osmv_query_req_t * const p_query_req)
{
ib_api_status_t status;
ib_mad_t *p_mad_hdr;
ib_sa_mad_t *p_sa_mad;
osm_madw_t *p_madw;
osm_log_t *p_log = p_bind->p_log;
static atomic32_t trans_id;
boolean_t sync;
osmv_query_req_t *p_query_req_copy;
OSM_LOG_ENTER(p_log);
/*
since the sm_lid might change we obtain it every send
(actually it is cached in the bind object and refreshed
every 30sec by this proc)
*/
if (time(NULL) > p_bind->last_lids_update_sec + 30) {
status = update_umad_port(p_bind->p_vendor);
if (status != IB_SUCCESS) {
OSM_LOG(p_log, OSM_LOG_ERROR, "ERR 5509: "
"Failed to obtain the SM lid\n");
goto Exit;
}
p_bind->last_lids_update_sec = time(NULL);
}
/* Get a MAD wrapper for the send */
p_madw = osm_mad_pool_get(p_bind->p_mad_pool,
p_bind->h_bind, MAD_BLOCK_SIZE, NULL);
if (p_madw == NULL) {
OSM_LOG(p_log, OSM_LOG_ERROR, "ERR 5510: "
"Unable to acquire MAD\n");
status = IB_INSUFFICIENT_RESOURCES;
goto Exit;
}
/* Initialize the Sent MAD: */
/* Initialize the MAD buffer for the send operation. */
p_mad_hdr = osm_madw_get_mad_ptr(p_madw);
p_sa_mad = osm_madw_get_sa_mad_ptr(p_madw);
/* Get a new transaction Id */
cl_atomic_inc(&trans_id);
/* Cleanup the MAD from any residue */
memset(p_sa_mad, 0, MAD_BLOCK_SIZE);
/* Initialize the standard MAD header. */
ib_mad_init_new(p_mad_hdr, /* mad pointer */
IB_MCLASS_SUBN_ADM, /* class */
(uint8_t) 2, /* version */
p_sa_mad_data->method, /* method */
cl_hton64((uint64_t) trans_id), /* tid */
p_sa_mad_data->attr_id, /* attr id */
p_sa_mad_data->attr_mod /* attr mod */);
/* Set the query information. */
p_sa_mad->sm_key = p_query_req->sm_key;
p_sa_mad->attr_offset = 0;
p_sa_mad->comp_mask = p_sa_mad_data->comp_mask;
#ifdef DUAL_SIDED_RMPP
if (p_sa_mad->method == IB_MAD_METHOD_GETMULTI)
p_sa_mad->rmpp_flags = IB_RMPP_FLAG_ACTIVE;
#endif
if (p_sa_mad->comp_mask) {
memcpy(p_sa_mad->data, p_sa_mad_data->p_attr,
ib_get_attr_size(p_sa_mad_data->attr_offset));
}
/*
Provide the address to send to
*/
p_madw->mad_addr.dest_lid =
cl_hton16(p_bind->p_vendor->umad_port.sm_lid);
p_madw->mad_addr.addr_type.smi.source_lid =
cl_hton16(p_bind->p_vendor->umad_port.base_lid);
p_madw->mad_addr.addr_type.gsi.remote_qp = CL_HTON32(1);
p_madw->resp_expected = TRUE;
p_madw->fail_msg = CL_DISP_MSGID_NONE;
/*
Provide MAD context such that the call back will know what to do.
We have to keep the entire request structure so we know the CB.
Since we can not rely on the client to keep it around until
the response - we duplicate it and will later dispose it (in CB).
To store on the MADW we cast it into what opensm has:
p_madw->context.ni_context.node_guid
*/
p_query_req_copy = malloc(sizeof(*p_query_req_copy));
*p_query_req_copy = *p_query_req;
p_madw->context.ni_context.node_guid =
(ib_net64_t) (long)p_query_req_copy;
/* we can support async as well as sync calls */
sync = ((p_query_req->flags & OSM_SA_FLAGS_SYNC) == OSM_SA_FLAGS_SYNC);
/* send the mad asynchronously */
status = osm_vendor_send(osm_madw_get_bind_handle(p_madw),
p_madw, p_madw->resp_expected);
/* if synchronous - wait on the event */
if (sync) {
OSM_LOG(p_log, OSM_LOG_DEBUG, "Waiting for async event\n");
cl_event_wait_on(&p_bind->sync_event, EVENT_NO_TIMEOUT, FALSE);
cl_event_reset(&p_bind->sync_event);
status = p_madw->status;
}
Exit:
OSM_LOG_EXIT(p_log);
return status;
}
static void
__osm_sa_pkey_by_comp_mask(IN osm_sa_t * sa,
IN const osm_port_t * const p_port,
osm_pkey_search_ctxt_t * const p_ctxt)
{
const ib_pkey_table_record_t *p_rcvd_rec;
ib_net64_t comp_mask;
osm_physp_t *p_physp;
uint8_t port_num;
uint8_t num_ports;
const osm_physp_t *p_req_physp;
OSM_LOG_ENTER(sa->p_log);
p_rcvd_rec = p_ctxt->p_rcvd_rec;
comp_mask = p_ctxt->comp_mask;
port_num = p_rcvd_rec->port_num;
p_req_physp = p_ctxt->p_req_physp;
/* if this is a switch port we can search all ports
otherwise we must be looking on port 0 */
if (p_port->p_node->node_info.node_type != IB_NODE_TYPE_SWITCH) {
/* we put it in the comp mask and port num */
port_num = p_port->p_physp->port_num;
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Using Physical Default Port Number: 0x%X (for End Node)\n",
port_num);
comp_mask |= IB_PKEY_COMPMASK_PORT;
}
if (comp_mask & IB_PKEY_COMPMASK_PORT) {
if (port_num < osm_node_get_num_physp(p_port->p_node)) {
p_physp =
osm_node_get_physp_ptr(p_port->p_node, port_num);
/* Check that the p_physp is valid, and that is shares a pkey
with the p_req_physp. */
if (p_physp &&
(osm_physp_share_pkey
(sa->p_log, p_req_physp, p_physp)))
__osm_sa_pkey_check_physp(sa, p_physp,
p_ctxt);
} else {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4603: "
"Given Physical Port Number: 0x%X is out of range should be < 0x%X\n",
port_num,
osm_node_get_num_physp(p_port->p_node));
goto Exit;
}
} else {
num_ports = osm_node_get_num_physp(p_port->p_node);
for (port_num = 0; port_num < num_ports; port_num++) {
p_physp =
osm_node_get_physp_ptr(p_port->p_node, port_num);
if (!p_physp)
continue;
/* if the requester and the p_physp don't share a pkey -
continue */
if (!osm_physp_share_pkey
(sa->p_log, p_req_physp, p_physp))
continue;
__osm_sa_pkey_check_physp(sa, p_physp, p_ctxt);
}
}
Exit:
OSM_LOG_EXIT(sa->p_log);
}
/*
Call back on new mad received:
We basically only need to set the context of the query.
Or report an error.
A pointer to the actual context of the request (a copy of the oriignal
request structure) is attached as the p_madw->context.ni_context.node_guid
*/
static void
__osmv_sa_mad_rcv_cb(IN osm_madw_t * p_madw,
IN void *bind_context, IN osm_madw_t * p_req_madw)
{
osmv_sa_bind_info_t *p_bind = (osmv_sa_bind_info_t *) bind_context;
osmv_query_req_t *p_query_req_copy = NULL;
osmv_query_res_t query_res;
ib_sa_mad_t *p_sa_mad;
ib_net16_t mad_status;
OSM_LOG_ENTER(p_bind->p_log);
if (!p_req_madw) {
OSM_LOG(p_bind->p_log, OSM_LOG_DEBUG,
"Ignoring a non-response mad\n");
osm_mad_pool_put(p_bind->p_mad_pool, p_madw);
goto Exit;
}
/* obtain the sent context since we store it during send in the ni_ctx */
p_query_req_copy =
(osmv_query_req_t *) (long *)(long)(p_req_madw->context.ni_context.
node_guid);
/* provide the context of the original request in the result */
query_res.query_context = p_query_req_copy->query_context;
/* provide the resulting madw */
query_res.p_result_madw = p_madw;
/* update the req fields */
p_sa_mad = (ib_sa_mad_t *) p_madw->p_mad;
/* if we got a remote error track it in the status */
mad_status = (ib_net16_t) (p_sa_mad->status & IB_SMP_STATUS_MASK);
if (mad_status != IB_SUCCESS) {
OSM_LOG(p_bind->p_log, OSM_LOG_ERROR, "ERR 5501: "
"Remote error:0x%04X\n", cl_ntoh16(mad_status));
query_res.status = IB_REMOTE_ERROR;
} else
query_res.status = IB_SUCCESS;
/* what if we have got back an empty mad ? */
if (!p_madw->mad_size) {
OSM_LOG(p_bind->p_log, OSM_LOG_ERROR, "ERR 5502: "
"Got an empty mad\n");
query_res.status = IB_ERROR;
}
if (IB_SUCCESS == mad_status) {
/* if we are in not in a method response of an rmpp nature we must get only 1 */
/* HACK: in the future we might need to be smarter for other methods... */
if (p_sa_mad->method != IB_MAD_METHOD_GETTABLE_RESP) {
query_res.result_cnt = 1;
} else {
#ifndef VENDOR_RMPP_SUPPORT
if (mad_status != IB_SUCCESS)
query_res.result_cnt = 0;
else
query_res.result_cnt = 1;
#else
if (ib_get_attr_size(p_sa_mad->attr_offset)) {
/* we used the offset value to calculate the
number of records in here */
query_res.result_cnt = (uintn_t)
((p_madw->mad_size - IB_SA_MAD_HDR_SIZE) /
ib_get_attr_size(p_sa_mad->attr_offset));
OSM_LOG(p_bind->p_log, OSM_LOG_DEBUG,
"Count = %u = %zu / %u (%zu)\n",
query_res.result_cnt,
p_madw->mad_size - IB_SA_MAD_HDR_SIZE,
ib_get_attr_size(p_sa_mad->attr_offset),
(p_madw->mad_size -
IB_SA_MAD_HDR_SIZE) %
ib_get_attr_size(p_sa_mad->attr_offset));
} else
query_res.result_cnt = 0;
#endif
}
}
query_res.query_type = p_query_req_copy->query_type;
p_query_req_copy->pfn_query_cb(&query_res);
if ((p_query_req_copy->flags & OSM_SA_FLAGS_SYNC) == OSM_SA_FLAGS_SYNC)
cl_event_signal(&p_bind->sync_event);
Exit:
/* free the copied query request if found */
if (p_query_req_copy)
free(p_query_req_copy);
/* put back the request madw */
if (p_req_madw)
osm_mad_pool_put(p_bind->p_mad_pool, p_req_madw);
OSM_LOG_EXIT(p_bind->p_log);
}
void osm_pkey_rec_rcv_process(IN void *ctx, IN void *data)
{
osm_sa_t *sa = ctx;
osm_madw_t *p_madw = data;
const ib_sa_mad_t *p_rcvd_mad;
const ib_pkey_table_record_t *p_rcvd_rec;
const osm_port_t *p_port = NULL;
const ib_pkey_table_t *p_pkey;
cl_qlist_t rec_list;
osm_pkey_search_ctxt_t context;
ib_api_status_t status = IB_SUCCESS;
ib_net64_t comp_mask;
osm_physp_t *p_req_physp;
CL_ASSERT(sa);
OSM_LOG_ENTER(sa->p_log);
CL_ASSERT(p_madw);
p_rcvd_mad = osm_madw_get_sa_mad_ptr(p_madw);
p_rcvd_rec =
(ib_pkey_table_record_t *) ib_sa_mad_get_payload_ptr(p_rcvd_mad);
comp_mask = p_rcvd_mad->comp_mask;
CL_ASSERT(p_rcvd_mad->attr_id == IB_MAD_ATTR_PKEY_TBL_RECORD);
/* we only support SubnAdmGet and SubnAdmGetTable methods */
if (p_rcvd_mad->method != IB_MAD_METHOD_GET &&
p_rcvd_mad->method != IB_MAD_METHOD_GETTABLE) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4605: "
"Unsupported Method (%s)\n",
ib_get_sa_method_str(p_rcvd_mad->method));
osm_sa_send_error(sa, p_madw, IB_MAD_STATUS_UNSUP_METHOD_ATTR);
goto Exit;
}
/*
p922 - P_KeyTableRecords shall only be provided in response
to trusted requests.
Check that the requester is a trusted one.
*/
if (p_rcvd_mad->sm_key != sa->p_subn->opt.sa_key) {
/* This is not a trusted requester! */
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4608: "
"Request from non-trusted requester: "
"Given SM_Key:0x%016" PRIx64 "\n",
cl_ntoh64(p_rcvd_mad->sm_key));
osm_sa_send_error(sa, p_madw, IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
/* update the requester physical port. */
p_req_physp = osm_get_physp_by_mad_addr(sa->p_log, sa->p_subn,
osm_madw_get_mad_addr_ptr
(p_madw));
if (p_req_physp == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4604: "
"Cannot find requester physical port\n");
goto Exit;
}
p_pkey = (ib_pkey_table_t *) ib_sa_mad_get_payload_ptr(p_rcvd_mad);
cl_qlist_init(&rec_list);
context.p_rcvd_rec = p_rcvd_rec;
context.p_list = &rec_list;
context.comp_mask = p_rcvd_mad->comp_mask;
context.sa = sa;
context.block_num = p_rcvd_rec->block_num;
context.p_req_physp = p_req_physp;
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Got Query Lid:%u(%02X), Block:0x%02X(%02X), Port:0x%02X(%02X)\n",
cl_ntoh16(p_rcvd_rec->lid),
(comp_mask & IB_PKEY_COMPMASK_LID) != 0, p_rcvd_rec->port_num,
(comp_mask & IB_PKEY_COMPMASK_PORT) != 0, p_rcvd_rec->block_num,
(comp_mask & IB_PKEY_COMPMASK_BLOCK) != 0);
cl_plock_acquire(sa->p_lock);
/*
If the user specified a LID, it obviously narrows our
work load, since we don't have to search every port
*/
if (comp_mask & IB_PKEY_COMPMASK_LID) {
status = osm_get_port_by_base_lid(sa->p_subn, p_rcvd_rec->lid,
&p_port);
if (status != IB_SUCCESS || p_port == NULL) {
status = IB_NOT_FOUND;
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 460B: "
"No port found with LID %u\n",
cl_ntoh16(p_rcvd_rec->lid));
}
}
if (status == IB_SUCCESS) {
/* if we got a unique port - no need for a port search */
if (p_port)
/* this does the loop on all the port phys ports */
__osm_sa_pkey_by_comp_mask(sa, p_port, &context);
else
cl_qmap_apply_func(&sa->p_subn->port_guid_tbl,
__osm_sa_pkey_by_comp_mask_cb,
&context);
}
cl_plock_release(sa->p_lock);
osm_sa_respond(sa, p_madw, sizeof(ib_pkey_table_record_t), &rec_list);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
void osm_slvl_rec_rcv_process(IN void *ctx, IN void *data)
{
osm_sa_t *sa = ctx;
osm_madw_t *p_madw = data;
const ib_sa_mad_t *p_rcvd_mad;
const ib_slvl_table_record_t *p_rcvd_rec;
const osm_port_t *p_port = NULL;
cl_qlist_t rec_list;
osm_slvl_search_ctxt_t context;
ib_api_status_t status = IB_SUCCESS;
ib_net64_t comp_mask;
osm_physp_t *p_req_physp;
CL_ASSERT(sa);
OSM_LOG_ENTER(sa->p_log);
CL_ASSERT(p_madw);
p_rcvd_mad = osm_madw_get_sa_mad_ptr(p_madw);
p_rcvd_rec =
(ib_slvl_table_record_t *) ib_sa_mad_get_payload_ptr(p_rcvd_mad);
comp_mask = p_rcvd_mad->comp_mask;
CL_ASSERT(p_rcvd_mad->attr_id == IB_MAD_ATTR_SLVL_RECORD);
/* we only support SubnAdmGet and SubnAdmGetTable methods */
if (p_rcvd_mad->method != IB_MAD_METHOD_GET &&
p_rcvd_mad->method != IB_MAD_METHOD_GETTABLE) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 2604: "
"Unsupported Method (%s)\n",
ib_get_sa_method_str(p_rcvd_mad->method));
osm_sa_send_error(sa, p_madw, IB_MAD_STATUS_UNSUP_METHOD_ATTR);
goto Exit;
}
/* update the requester physical port. */
p_req_physp = osm_get_physp_by_mad_addr(sa->p_log, sa->p_subn,
osm_madw_get_mad_addr_ptr
(p_madw));
if (p_req_physp == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 2603: "
"Cannot find requester physical port\n");
goto Exit;
}
cl_qlist_init(&rec_list);
context.p_rcvd_rec = p_rcvd_rec;
context.p_list = &rec_list;
context.comp_mask = p_rcvd_mad->comp_mask;
context.sa = sa;
context.in_port_num = p_rcvd_rec->in_port_num;
context.p_req_physp = p_req_physp;
cl_plock_acquire(sa->p_lock);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Got Query Lid:%u(%02X), In-Port:0x%02X(%02X), Out-Port:0x%02X(%02X)\n",
cl_ntoh16(p_rcvd_rec->lid),
(comp_mask & IB_SLVL_COMPMASK_LID) != 0,
p_rcvd_rec->in_port_num,
(comp_mask & IB_SLVL_COMPMASK_IN_PORT) != 0,
p_rcvd_rec->out_port_num,
(comp_mask & IB_SLVL_COMPMASK_OUT_PORT) != 0);
/*
If the user specified a LID, it obviously narrows our
work load, since we don't have to search every port
*/
if (comp_mask & IB_SLVL_COMPMASK_LID) {
p_port = osm_get_port_by_lid(sa->p_subn, p_rcvd_rec->lid);
if (!p_port) {
status = IB_NOT_FOUND;
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 2608: "
"No port found with LID %u\n",
cl_ntoh16(p_rcvd_rec->lid));
}
}
if (status == IB_SUCCESS) {
/* if we have a unique port - no need for a port search */
if (p_port)
/* this does the loop on all the port phys ports */
sa_slvl_by_comp_mask(sa, p_port, &context);
else
cl_qmap_apply_func(&sa->p_subn->port_guid_tbl,
sa_slvl_by_comp_mask_cb, &context);
}
cl_plock_release(sa->p_lock);
osm_sa_respond(sa, p_madw, sizeof(ib_slvl_table_record_t), &rec_list);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
static void
__osm_sa_inform_info_rec_by_comp_mask(IN osm_sa_t * sa,
IN const osm_infr_t * const p_infr,
osm_iir_search_ctxt_t * const p_ctxt)
{
const ib_inform_info_record_t *p_rcvd_rec = NULL;
ib_net64_t comp_mask;
ib_net64_t portguid;
osm_port_t *p_subscriber_port;
osm_physp_t *p_subscriber_physp;
const osm_physp_t *p_req_physp;
osm_iir_item_t *p_rec_item;
OSM_LOG_ENTER(sa->p_log);
p_rcvd_rec = p_ctxt->p_rcvd_rec;
comp_mask = p_ctxt->comp_mask;
p_req_physp = p_ctxt->p_req_physp;
if (comp_mask & IB_IIR_COMPMASK_SUBSCRIBERGID &&
memcmp(&p_infr->inform_record.subscriber_gid,
&p_ctxt->subscriber_gid,
sizeof(p_infr->inform_record.subscriber_gid)))
goto Exit;
if (comp_mask & IB_IIR_COMPMASK_ENUM &&
p_infr->inform_record.subscriber_enum != p_ctxt->subscriber_enum)
goto Exit;
/* Implement any other needed search cases */
/* Ensure pkey is shared before returning any records */
portguid = p_infr->inform_record.subscriber_gid.unicast.interface_id;
p_subscriber_port = osm_get_port_by_guid(sa->p_subn, portguid);
if (p_subscriber_port == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 430D: "
"Invalid subscriber port guid: 0x%016" PRIx64 "\n",
cl_ntoh64(portguid));
goto Exit;
}
/* get the subscriber InformInfo physical port */
p_subscriber_physp = p_subscriber_port->p_physp;
/* make sure that the requester and subscriber port can access each other
according to the current partitioning. */
if (!osm_physp_share_pkey
(sa->p_log, p_req_physp, p_subscriber_physp)) {
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"requester and subscriber ports don't share pkey\n");
goto Exit;
}
p_rec_item = malloc(sizeof(*p_rec_item));
if (p_rec_item == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 430E: "
"rec_item alloc failed\n");
goto Exit;
}
memcpy((void *)&p_rec_item->rec, (void *)&p_infr->inform_record,
sizeof(ib_inform_info_record_t));
cl_qlist_insert_tail(p_ctxt->p_list, &p_rec_item->list_item);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
/****f* opensm: SM/__osm_sm_mad_ctrl_process_get_resp
* NAME
* __osm_sm_mad_ctrl_process_get_resp
*
* DESCRIPTION
* This function handles method GetResp() for received MADs.
* This is the most common path for QP0 MADs.
*
* SYNOPSIS
*/
static void
__osm_sm_mad_ctrl_process_get_resp(IN osm_sm_mad_ctrl_t * const p_ctrl,
IN osm_madw_t * p_madw,
IN void *transaction_context)
{
ib_smp_t *p_smp;
cl_status_t status;
osm_madw_t *p_old_madw;
cl_disp_msgid_t msg_id = CL_DISP_MSGID_NONE;
OSM_LOG_ENTER(p_ctrl->p_log);
CL_ASSERT(p_madw);
CL_ASSERT(transaction_context);
p_smp = osm_madw_get_smp_ptr(p_madw);
if (p_smp->mgmt_class == IB_MCLASS_SUBN_DIR && !ib_smp_is_d(p_smp)) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3102: "
"'D' bit not set in returned SMP\n");
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_ERROR);
}
p_old_madw = (osm_madw_t *) transaction_context;
__osm_sm_mad_ctrl_update_wire_stats(p_ctrl);
/*
Copy the MAD Wrapper context from the requesting MAD
to the new MAD. This mechanism allows the recipient
controller to recover its own context regarding this
MAD transaction. Once we've copied the context, we
can return the original MAD to the pool.
*/
osm_madw_copy_context(p_madw, p_old_madw);
osm_mad_pool_put(p_ctrl->p_mad_pool, p_old_madw);
/*
Note that attr_id (like the rest of the MAD) is in
network byte order.
*/
switch (p_smp->attr_id) {
case IB_MAD_ATTR_NODE_DESC:
msg_id = OSM_MSG_MAD_NODE_DESC;
break;
case IB_MAD_ATTR_NODE_INFO:
msg_id = OSM_MSG_MAD_NODE_INFO;
break;
case IB_MAD_ATTR_SWITCH_INFO:
msg_id = OSM_MSG_MAD_SWITCH_INFO;
break;
case IB_MAD_ATTR_PORT_INFO:
msg_id = OSM_MSG_MAD_PORT_INFO;
break;
case IB_MAD_ATTR_LIN_FWD_TBL:
msg_id = OSM_MSG_MAD_LFT;
break;
case IB_MAD_ATTR_MCAST_FWD_TBL:
msg_id = OSM_MSG_MAD_MFT;
break;
case IB_MAD_ATTR_SM_INFO:
msg_id = OSM_MSG_MAD_SM_INFO;
break;
case IB_MAD_ATTR_SLVL_TABLE:
msg_id = OSM_MSG_MAD_SLVL;
break;
case IB_MAD_ATTR_VL_ARBITRATION:
msg_id = OSM_MSG_MAD_VL_ARB;
break;
case IB_MAD_ATTR_P_KEY_TABLE:
msg_id = OSM_MSG_MAD_PKEY;
break;
case IB_MAD_ATTR_GUID_INFO:
case IB_MAD_ATTR_CLASS_PORT_INFO:
case IB_MAD_ATTR_NOTICE:
case IB_MAD_ATTR_INFORM_INFO:
default:
cl_atomic_inc(&p_ctrl->p_stats->qp0_mads_rcvd_unknown);
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3103: "
"Unsupported attribute = 0x%X\n",
cl_ntoh16(p_smp->attr_id));
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_ERROR);
goto Exit;
}
if (msg_id == CL_DISP_MSGID_NONE)
goto Exit;
/*
Post this MAD to the dispatcher for asynchronous
processing by the appropriate controller.
*/
OSM_LOG(p_ctrl->p_log, OSM_LOG_DEBUG, "Posting Dispatcher message %s\n",
osm_get_disp_msg_str(msg_id));
status = cl_disp_post(p_ctrl->h_disp, msg_id, p_madw,
__osm_sm_mad_ctrl_disp_done_callback, p_ctrl);
if (status != CL_SUCCESS) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3104: "
"Dispatcher post message failed (%s) for attribute = 0x%X\n",
CL_STATUS_MSG(status), cl_ntoh16(p_smp->attr_id));
goto Exit;
}
Exit:
OSM_LOG_EXIT(p_ctrl->p_log);
}
/**********************************************************************
Received a Get(InformInfoRecord) or GetTable(InformInfoRecord) MAD
**********************************************************************/
static void
osm_infr_rcv_process_get_method(IN osm_sa_t * sa,
IN osm_madw_t * const p_madw)
{
char gid_str[INET6_ADDRSTRLEN];
ib_sa_mad_t *p_rcvd_mad;
const ib_inform_info_record_t *p_rcvd_rec;
cl_qlist_t rec_list;
osm_iir_search_ctxt_t context;
osm_physp_t *p_req_physp;
osm_iir_item_t *item;
OSM_LOG_ENTER(sa->p_log);
CL_ASSERT(p_madw);
p_rcvd_mad = osm_madw_get_sa_mad_ptr(p_madw);
p_rcvd_rec =
(ib_inform_info_record_t *) ib_sa_mad_get_payload_ptr(p_rcvd_mad);
/* update the requester physical port. */
p_req_physp = osm_get_physp_by_mad_addr(sa->p_log, sa->p_subn,
osm_madw_get_mad_addr_ptr
(p_madw));
if (p_req_physp == NULL) {
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4309: "
"Cannot find requester physical port\n");
goto Exit;
}
if (osm_log_is_active(sa->p_log, OSM_LOG_DEBUG))
osm_dump_inform_info_record(sa->p_log, p_rcvd_rec,
OSM_LOG_DEBUG);
cl_qlist_init(&rec_list);
context.p_rcvd_rec = p_rcvd_rec;
context.p_list = &rec_list;
context.comp_mask = p_rcvd_mad->comp_mask;
context.subscriber_gid = p_rcvd_rec->subscriber_gid;
context.subscriber_enum = p_rcvd_rec->subscriber_enum;
context.sa = sa;
context.p_req_physp = p_req_physp;
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Query Subscriber GID:%s(%02X) Enum:0x%X(%02X)\n",
inet_ntop(AF_INET6, p_rcvd_rec->subscriber_gid.raw,
gid_str, sizeof gid_str),
(p_rcvd_mad->comp_mask & IB_IIR_COMPMASK_SUBSCRIBERGID) != 0,
cl_ntoh16(p_rcvd_rec->subscriber_enum),
(p_rcvd_mad->comp_mask & IB_IIR_COMPMASK_ENUM) != 0);
cl_plock_acquire(sa->p_lock);
cl_qlist_apply_func(&sa->p_subn->sa_infr_list,
__osm_sa_inform_info_rec_by_comp_mask_cb, &context);
cl_plock_release(sa->p_lock);
/* clear reserved and pad fields in InformInfoRecord */
for (item = (osm_iir_item_t *) cl_qlist_head(&rec_list);
item != (osm_iir_item_t *) cl_qlist_end(&rec_list);
item = (osm_iir_item_t *)cl_qlist_next(&item->list_item)) {
memset(item->rec.reserved, 0, sizeof(item->rec.reserved));
memset(item->rec.pad, 0, sizeof(item->rec.pad));
}
osm_sa_respond(sa, p_madw, sizeof(ib_inform_info_record_t), &rec_list);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
/****f* opensm: SM/__osm_sm_mad_ctrl_rcv_callback
* NAME
* __osm_sm_mad_ctrl_rcv_callback
*
* DESCRIPTION
* This is the callback from the transport layer for received MADs.
*
* SYNOPSIS
*/
static void
__osm_sm_mad_ctrl_rcv_callback(IN osm_madw_t * p_madw,
IN void *bind_context,
IN osm_madw_t * p_req_madw)
{
osm_sm_mad_ctrl_t *p_ctrl = (osm_sm_mad_ctrl_t *) bind_context;
ib_smp_t *p_smp;
ib_net16_t status;
OSM_LOG_ENTER(p_ctrl->p_log);
CL_ASSERT(p_madw);
/*
A MAD was received from the wire, possibly in response to a request.
*/
cl_atomic_inc(&p_ctrl->p_stats->qp0_mads_rcvd);
OSM_LOG(p_ctrl->p_log, OSM_LOG_DEBUG, "%u QP0 MADs received\n",
p_ctrl->p_stats->qp0_mads_rcvd);
p_smp = osm_madw_get_smp_ptr(p_madw);
/* if we are closing down simply do nothing */
if (osm_exit_flag) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR,
"Ignoring received mad - since we are exiting\n");
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_DEBUG);
/* retire the mad or put it back */
if (ib_smp_is_response(p_smp) ||
(p_smp->method == IB_MAD_METHOD_TRAP_REPRESS)) {
CL_ASSERT(p_madw->resp_expected == FALSE);
__osm_sm_mad_ctrl_retire_trans_mad(p_ctrl, p_madw);
} else if (p_madw->resp_expected == TRUE)
__osm_sm_mad_ctrl_retire_trans_mad(p_ctrl, p_madw);
else
osm_mad_pool_put(p_ctrl->p_mad_pool, p_madw);
goto Exit;
}
if (osm_log_is_active(p_ctrl->p_log, OSM_LOG_FRAMES))
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_FRAMES);
if (p_smp->mgmt_class == IB_MCLASS_SUBN_DIR)
status = ib_smp_get_status(p_smp);
else
status = p_smp->status;
if (status != 0) {
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3111: "
"Error status = 0x%X\n", status);
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_ERROR);
}
switch (p_smp->method) {
case IB_MAD_METHOD_GET_RESP:
CL_ASSERT(p_req_madw != NULL);
__osm_sm_mad_ctrl_process_get_resp(p_ctrl, p_madw, p_req_madw);
break;
case IB_MAD_METHOD_GET:
CL_ASSERT(p_req_madw == NULL);
__osm_sm_mad_ctrl_process_get(p_ctrl, p_madw);
break;
case IB_MAD_METHOD_TRAP:
CL_ASSERT(p_req_madw == NULL);
__osm_sm_mad_ctrl_process_trap(p_ctrl, p_madw);
break;
case IB_MAD_METHOD_SET:
CL_ASSERT(p_req_madw == NULL);
__osm_sm_mad_ctrl_process_set(p_ctrl, p_madw);
break;
case IB_MAD_METHOD_SEND:
case IB_MAD_METHOD_REPORT:
case IB_MAD_METHOD_REPORT_RESP:
case IB_MAD_METHOD_TRAP_REPRESS:
default:
cl_atomic_inc(&p_ctrl->p_stats->qp0_mads_rcvd_unknown);
OSM_LOG(p_ctrl->p_log, OSM_LOG_ERROR, "ERR 3112: "
"Unsupported method = 0x%X\n", p_smp->method);
osm_dump_dr_smp(p_ctrl->p_log, p_smp, OSM_LOG_ERROR);
osm_mad_pool_put(p_ctrl->p_mad_pool, p_madw);
goto Exit;
}
Exit:
OSM_LOG_EXIT(p_ctrl->p_log);
}
/*********************************************************************
Received a Set(InformInfo) MAD
**********************************************************************/
static void
osm_infr_rcv_process_set_method(IN osm_sa_t * sa,
IN osm_madw_t * const p_madw)
{
ib_sa_mad_t *p_sa_mad;
ib_inform_info_t *p_recvd_inform_info;
osm_infr_t inform_info_rec; /* actual inform record to be stored for reports */
osm_infr_t *p_infr;
ib_net32_t qpn;
uint8_t resp_time_val;
ib_api_status_t res;
OSM_LOG_ENTER(sa->p_log);
CL_ASSERT(p_madw);
p_sa_mad = osm_madw_get_sa_mad_ptr(p_madw);
p_recvd_inform_info =
(ib_inform_info_t *) ib_sa_mad_get_payload_ptr(p_sa_mad);
#if 0
if (osm_log_is_active(sa->p_log, OSM_LOG_DEBUG))
osm_dump_inform_info(sa->p_log, p_recvd_inform_info,
OSM_LOG_DEBUG);
#endif
/* Grab the lock */
cl_plock_excl_acquire(sa->p_lock);
/* define the inform record */
inform_info_rec.inform_record.inform_info = *p_recvd_inform_info;
/* following C13-32.1.2 Tbl 120: we only copy the source address vector */
inform_info_rec.report_addr = p_madw->mad_addr;
/* we will need to know the mad srvc to send back through */
inform_info_rec.h_bind = p_madw->h_bind;
inform_info_rec.sa = sa;
/* update the subscriber GID according to mad address */
res = osm_get_gid_by_mad_addr(sa->p_log, sa->p_subn, &p_madw->mad_addr,
&inform_info_rec.inform_record.
subscriber_gid);
if (res != IB_SUCCESS) {
cl_plock_release(sa->p_lock);
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4308 "
"Subscribe Request from unknown LID: %u\n",
cl_ntoh16(p_madw->mad_addr.dest_lid));
osm_sa_send_error(sa, p_madw, IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
/* HACK: enum is always 0 (currently) */
inform_info_rec.inform_record.subscriber_enum = 0;
/* Subscribe values above 1 are undefined */
if (p_recvd_inform_info->subscribe > 1) {
cl_plock_release(sa->p_lock);
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4308 "
"Invalid subscribe: %d\n",
p_recvd_inform_info->subscribe);
osm_sa_send_error(sa, p_madw, IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
/*
* MODIFICATIONS DONE ON INCOMING REQUEST:
*
* QPN:
* Internally we keep the QPN field of the InformInfo updated
* so we can simply compare it in the record - when finding such.
*/
if (p_recvd_inform_info->subscribe) {
ib_inform_info_set_qpn(&inform_info_rec.inform_record.
inform_info,
inform_info_rec.report_addr.addr_type.
gsi.remote_qp);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"Subscribe Request with QPN: 0x%06X\n",
cl_ntoh32(inform_info_rec.report_addr.addr_type.gsi.
remote_qp));
} else {
ib_inform_info_get_qpn_resp_time(p_recvd_inform_info->g_or_v.
generic.qpn_resp_time_val,
&qpn, &resp_time_val);
OSM_LOG(sa->p_log, OSM_LOG_DEBUG,
"UnSubscribe Request with QPN: 0x%06X\n",
cl_ntoh32(qpn));
}
/* If record exists with matching InformInfo */
p_infr =
osm_infr_get_by_rec(sa->p_subn, sa->p_log, &inform_info_rec);
/* check to see if the request was for subscribe */
if (p_recvd_inform_info->subscribe) {
/* validate the request for a new or update InformInfo */
if (__validate_infr(sa, &inform_info_rec) != TRUE) {
cl_plock_release(sa->p_lock);
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4305: "
"Failed to validate a new inform object\n");
/* o13-13.1.1: we need to set the subscribe bit to 0 */
p_recvd_inform_info->subscribe = 0;
osm_sa_send_error(sa, p_madw,
IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
}
/* ok - we can try and create a new entry */
if (p_infr == NULL) {
/* Create the instance of the osm_infr_t object */
p_infr = osm_infr_new(&inform_info_rec);
if (p_infr == NULL) {
cl_plock_release(sa->p_lock);
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4306: "
"Failed to create a new inform object\n");
/* o13-13.1.1: we need to set the subscribe bit to 0 */
p_recvd_inform_info->subscribe = 0;
osm_sa_send_error(sa, p_madw,
IB_SA_MAD_STATUS_NO_RESOURCES);
goto Exit;
}
/* Add this new osm_infr_t object to subnet object */
osm_infr_insert_to_db(sa->p_subn, sa->p_log, p_infr);
} else
/* Update the old instance of the osm_infr_t object */
p_infr->inform_record = inform_info_rec.inform_record;
/* We got an UnSubscribe request */
} else if (p_infr == NULL) {
cl_plock_release(sa->p_lock);
/* No Such Item - So Error */
OSM_LOG(sa->p_log, OSM_LOG_ERROR, "ERR 4307: "
"Failed to UnSubscribe to non existing inform object\n");
/* o13-13.1.1: we need to set the subscribe bit to 0 */
p_recvd_inform_info->subscribe = 0;
osm_sa_send_error(sa, p_madw, IB_SA_MAD_STATUS_REQ_INVALID);
goto Exit;
} else
/* Delete this object from the subnet list of informs */
osm_infr_remove_from_db(sa->p_subn, sa->p_log, p_infr);
cl_plock_release(sa->p_lock);
/* send the success response */
__osm_infr_rcv_respond(sa, p_madw);
Exit:
OSM_LOG_EXIT(sa->p_log);
}
static boolean_t
__osm_link_mgr_set_physp_pi(osm_sm_t * sm,
IN osm_physp_t * const p_physp,
IN uint8_t const port_state)
{
uint8_t payload[IB_SMP_DATA_SIZE];
ib_port_info_t *const p_pi = (ib_port_info_t *) payload;
const ib_port_info_t *p_old_pi;
osm_madw_context_t context;
osm_node_t *p_node;
ib_api_status_t status;
uint8_t port_num;
uint8_t mtu;
uint8_t op_vls;
boolean_t esp0 = FALSE;
boolean_t send_set = FALSE;
osm_physp_t *p_remote_physp;
OSM_LOG_ENTER(sm->p_log);
p_node = osm_physp_get_node_ptr(p_physp);
port_num = osm_physp_get_port_num(p_physp);
if (port_num == 0) {
/*
CAs don't have a port 0, and for switch port 0,
we need to check if this is enhanced or base port 0.
For base port 0 the following parameters are not valid (p822, table 145).
*/
if (!p_node->sw) {
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 4201: "
"Cannot find switch by guid: 0x%" PRIx64 "\n",
cl_ntoh64(p_node->node_info.node_guid));
goto Exit;
}
if (ib_switch_info_is_enhanced_port0(&p_node->sw->switch_info)
== FALSE) {
/* This means the switch doesn't support enhanced port 0.
Can skip it. */
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Skipping port 0, GUID 0x%016" PRIx64 "\n",
cl_ntoh64(osm_physp_get_port_guid(p_physp)));
goto Exit;
}
esp0 = TRUE;
}
/*
PAST THIS POINT WE ARE HANDLING EITHER A NON PORT 0 OR ENHANCED PORT 0
*/
p_old_pi = &p_physp->port_info;
memset(payload, 0, IB_SMP_DATA_SIZE);
memcpy(payload, p_old_pi, sizeof(ib_port_info_t));
/*
Should never write back a value that is bigger then 3 in
the PortPhysicalState field - so can not simply copy!
Actually we want to write there:
port physical state - no change,
link down default state = polling
port state - as requested.
*/
p_pi->state_info2 = 0x02;
ib_port_info_set_port_state(p_pi, port_state);
if (ib_port_info_get_link_down_def_state(p_pi) !=
ib_port_info_get_link_down_def_state(p_old_pi))
send_set = TRUE;
/* didn't get PortInfo before */
if (!ib_port_info_get_port_state(p_old_pi))
send_set = TRUE;
/* we only change port fields if we do not change state */
if (port_state == IB_LINK_NO_CHANGE) {
/* The following fields are relevant only for CA port, router, or Enh. SP0 */
if (osm_node_get_type(p_node) != IB_NODE_TYPE_SWITCH ||
port_num == 0) {
p_pi->m_key = sm->p_subn->opt.m_key;
if (memcmp(&p_pi->m_key, &p_old_pi->m_key,
sizeof(p_pi->m_key)))
send_set = TRUE;
p_pi->subnet_prefix = sm->p_subn->opt.subnet_prefix;
if (memcmp(&p_pi->subnet_prefix,
&p_old_pi->subnet_prefix,
sizeof(p_pi->subnet_prefix)))
send_set = TRUE;
p_pi->base_lid = osm_physp_get_base_lid(p_physp);
if (memcmp(&p_pi->base_lid, &p_old_pi->base_lid,
sizeof(p_pi->base_lid)))
send_set = TRUE;
/* we are initializing the ports with our local sm_base_lid */
p_pi->master_sm_base_lid = sm->p_subn->sm_base_lid;
if (memcmp(&p_pi->master_sm_base_lid,
&p_old_pi->master_sm_base_lid,
sizeof(p_pi->master_sm_base_lid)))
send_set = TRUE;
p_pi->m_key_lease_period =
sm->p_subn->opt.m_key_lease_period;
if (memcmp(&p_pi->m_key_lease_period,
&p_old_pi->m_key_lease_period,
sizeof(p_pi->m_key_lease_period)))
send_set = TRUE;
if (esp0 == FALSE)
p_pi->mkey_lmc = sm->p_subn->opt.lmc;
else {
if (sm->p_subn->opt.lmc_esp0)
p_pi->mkey_lmc = sm->p_subn->opt.lmc;
else
p_pi->mkey_lmc = 0;
}
if (memcmp(&p_pi->mkey_lmc, &p_old_pi->mkey_lmc,
sizeof(p_pi->mkey_lmc)))
send_set = TRUE;
ib_port_info_set_timeout(p_pi,
sm->p_subn->opt.
subnet_timeout);
if (ib_port_info_get_timeout(p_pi) !=
ib_port_info_get_timeout(p_old_pi))
send_set = TRUE;
}
/*
Several timeout mechanisms:
*/
p_remote_physp = osm_physp_get_remote(p_physp);
if (port_num != 0 && p_remote_physp) {
if (osm_node_get_type(osm_physp_get_node_ptr(p_physp))
== IB_NODE_TYPE_ROUTER) {
ib_port_info_set_hoq_lifetime(p_pi,
sm->p_subn->
opt.
leaf_head_of_queue_lifetime);
} else
if (osm_node_get_type
(osm_physp_get_node_ptr(p_physp)) ==
IB_NODE_TYPE_SWITCH) {
/* Is remote end CA or router (a leaf port) ? */
if (osm_node_get_type
(osm_physp_get_node_ptr(p_remote_physp)) !=
IB_NODE_TYPE_SWITCH) {
ib_port_info_set_hoq_lifetime(p_pi,
sm->
p_subn->
opt.
leaf_head_of_queue_lifetime);
ib_port_info_set_vl_stall_count(p_pi,
sm->
p_subn->
opt.
leaf_vl_stall_count);
} else {
ib_port_info_set_hoq_lifetime(p_pi,
sm->
p_subn->
opt.
head_of_queue_lifetime);
ib_port_info_set_vl_stall_count(p_pi,
sm->
p_subn->
opt.
vl_stall_count);
}
}
if (ib_port_info_get_hoq_lifetime(p_pi) !=
ib_port_info_get_hoq_lifetime(p_old_pi) ||
ib_port_info_get_vl_stall_count(p_pi) !=
ib_port_info_get_vl_stall_count(p_old_pi))
send_set = TRUE;
}
ib_port_info_set_phy_and_overrun_err_thd(p_pi,
sm->p_subn->opt.
local_phy_errors_threshold,
sm->p_subn->opt.
overrun_errors_threshold);
if (memcmp(&p_pi->error_threshold, &p_old_pi->error_threshold,
sizeof(p_pi->error_threshold)))
send_set = TRUE;
/*
Set the easy common parameters for all port types,
then determine the neighbor MTU.
*/
p_pi->link_width_enabled = p_old_pi->link_width_supported;
if (memcmp(&p_pi->link_width_enabled,
&p_old_pi->link_width_enabled,
sizeof(p_pi->link_width_enabled)))
send_set = TRUE;
if (sm->p_subn->opt.force_link_speed &&
(sm->p_subn->opt.force_link_speed != 15 ||
ib_port_info_get_link_speed_enabled(p_pi) !=
ib_port_info_get_link_speed_sup(p_pi))) {
ib_port_info_set_link_speed_enabled(p_pi,
sm->p_subn->opt.
force_link_speed);
if (memcmp(&p_pi->link_speed, &p_old_pi->link_speed,
sizeof(p_pi->link_speed)))
send_set = TRUE;
}
/* calc new op_vls and mtu */
op_vls =
osm_physp_calc_link_op_vls(sm->p_log, sm->p_subn, p_physp);
mtu = osm_physp_calc_link_mtu(sm->p_log, p_physp);
ib_port_info_set_neighbor_mtu(p_pi, mtu);
if (ib_port_info_get_neighbor_mtu(p_pi) !=
ib_port_info_get_neighbor_mtu(p_old_pi))
send_set = TRUE;
ib_port_info_set_op_vls(p_pi, op_vls);
if (ib_port_info_get_op_vls(p_pi) !=
ib_port_info_get_op_vls(p_old_pi))
send_set = TRUE;
/* provide the vl_high_limit from the qos mgr */
if (sm->p_subn->opt.qos &&
p_physp->vl_high_limit != p_old_pi->vl_high_limit) {
send_set = TRUE;
p_pi->vl_high_limit = p_physp->vl_high_limit;
}
}
if (port_state != IB_LINK_NO_CHANGE &&
port_state != ib_port_info_get_port_state(p_old_pi)) {
send_set = TRUE;
if (port_state == IB_LINK_ACTIVE)
context.pi_context.active_transition = TRUE;
else
context.pi_context.active_transition = FALSE;
}
context.pi_context.node_guid = osm_node_get_node_guid(p_node);
context.pi_context.port_guid = osm_physp_get_port_guid(p_physp);
context.pi_context.set_method = TRUE;
context.pi_context.light_sweep = FALSE;
/* We need to send the PortInfoSet request with the new sm_lid
in the following cases:
1. There is a change in the values (send_set == TRUE)
2. This is a switch external port (so it wasn't handled yet by
osm_lid_mgr) and first_time_master_sweep flag on the subnet is TRUE,
which means the SM just became master, and it then needs to send at
PortInfoSet to every port.
*/
if (osm_node_get_type(p_node) == IB_NODE_TYPE_SWITCH && port_num
&& sm->p_subn->first_time_master_sweep == TRUE)
send_set = TRUE;
if (send_set)
status = osm_req_set(sm, osm_physp_get_dr_path_ptr(p_physp),
payload, sizeof(payload),
IB_MAD_ATTR_PORT_INFO,
cl_hton32(port_num),
CL_DISP_MSGID_NONE, &context);
Exit:
OSM_LOG_EXIT(sm->p_log);
return send_set;
}
/**********************************************************************
* Send SubnSet(SMInfo) SMP with HANDOVER attribute to the
* remote_sm indicated.
**********************************************************************/
static void state_mgr_send_handover(IN osm_sm_t * sm, IN osm_remote_sm_t * p_sm)
{
uint8_t payload[IB_SMP_DATA_SIZE];
ib_sm_info_t *p_smi = (ib_sm_info_t *) payload;
osm_madw_context_t context;
const osm_port_t *p_port;
ib_api_status_t status;
OSM_LOG_ENTER(sm->p_log);
/*
* Send a query of SubnSet(SMInfo) HANDOVER to the remote sm given.
*/
memset(&context, 0, sizeof(context));
p_port = p_sm->p_port;
if (p_port == NULL) {
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3316: "
"No port object on given remote_sm object\n");
goto Exit;
}
/* update the master_guid in the sm_state_mgr object according to */
/* the guid of the port where the new Master SM should reside. */
OSM_LOG(sm->p_log, OSM_LOG_VERBOSE,
"Handing over mastership. Updating sm_state_mgr master_guid: %016"
PRIx64 " (node %s)\n", cl_ntoh64(p_port->guid),
p_port->p_node ? p_port->p_node->print_desc : "UNKNOWN");
sm->master_sm_guid = p_port->guid;
context.smi_context.port_guid = p_port->guid;
context.smi_context.set_method = TRUE;
p_smi->guid = sm->p_subn->sm_port_guid;
p_smi->act_count = cl_hton32(sm->p_subn->p_osm->stats.qp0_mads_sent);
p_smi->pri_state = (uint8_t) (sm->p_subn->sm_state |
sm->p_subn->opt.sm_priority << 4);
/*
* Return 0 for the SM key unless we authenticate the requester
* as the master SM.
*/
if (ib_sminfo_get_state(&p_sm->smi) == IB_SMINFO_STATE_MASTER) {
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Responding to master SM with real sm_key\n");
p_smi->sm_key = sm->p_subn->opt.sm_key;
} else {
/* The requester is not authenticated as master - set sm_key to zero */
OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
"Responding to SM not master with zero sm_key\n");
p_smi->sm_key = 0;
}
status = osm_req_set(sm, osm_physp_get_dr_path_ptr(p_port->p_physp),
payload, sizeof(payload), IB_MAD_ATTR_SM_INFO,
IB_SMINFO_ATTR_MOD_HANDOVER, CL_DISP_MSGID_NONE,
&context);
if (status != IB_SUCCESS)
OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3317: "
"Failure requesting SMInfo (%s)\n",
ib_get_err_str(status));
Exit:
OSM_LOG_EXIT(sm->p_log);
}
