[ofa-general] Re: [PATCH RFC] opensm: mcast mgr improvements
Hal Rosenstock
hrosenstock at xsigo.com
Sun Dec 30 08:38:30 PST 2007
On Sat, 2007-12-29 at 18:27 +0000, Sasha Khapyorsky wrote:
> This improves handling of mcast join/leave requests storming. Now mcast
> routing will be recalculated for all mcast groups where changes occurred
> and not one by one. For this it queues mcast groups instead of mcast
> rerouting requests, this also makes state_mgr idle queue obsolete.
Looks like a nice improvement.
What testing has been done with this change ? Can you comment on any
results ?
For which branches is this change being proposed ?
-- Hal
> Signed-off-by: Sasha Khapyorsky <sashak at voltaire.com>
> ---
>
> Hi Yevgeny,
>
> For me it looks that it should solve the original problem (mcast group
> list is purged in osm_mcast_mgr_process()). Could you review and ideally
> test it? Thanks.
>
> Sasha
>
> ---
> opensm/include/opensm/osm_mcast_mgr.h | 14 +--
> opensm/include/opensm/osm_multicast.h | 2 +
> opensm/include/opensm/osm_sm.h | 2 +
> opensm/include/opensm/osm_state_mgr.h | 95 -----------------
> opensm/opensm/osm_mcast_mgr.c | 187 +++++++++++++++------------------
> opensm/opensm/osm_sm.c | 70 ++++++-------
> opensm/opensm/osm_state_mgr.c | 138 +------------------------
> 7 files changed, 130 insertions(+), 378 deletions(-)
>
> diff --git a/opensm/include/opensm/osm_mcast_mgr.h b/opensm/include/opensm/osm_mcast_mgr.h
> index 3e0b761..47b67ed 100644
> --- a/opensm/include/opensm/osm_mcast_mgr.h
> +++ b/opensm/include/opensm/osm_mcast_mgr.h
> @@ -100,7 +100,6 @@ typedef struct _osm_mcast_mgr {
> osm_req_t *p_req;
> osm_log_t *p_log;
> cl_plock_t *p_lock;
> -
> } osm_mcast_mgr_t;
> /*
> * FIELDS
> @@ -253,25 +252,22 @@ osm_signal_t osm_mcast_mgr_process(IN osm_mcast_mgr_t * const p_mgr);
> * Multicast Manager, Node Info Response Controller
> *********/
>
> -/****f* OpenSM: Multicast Manager/osm_mcast_mgr_process_mgrp_cb
> +/****f* OpenSM: Multicast Manager/osm_mcast_mgr_process_mgroups
> * NAME
> -* osm_mcast_mgr_process_mgrp_cb
> +* osm_mcast_mgr_process_mgroups
> *
> * DESCRIPTION
> -* Callback entry point for the osm_mcast_mgr_process_mgrp function.
> +* Process only requested mcast groups.
> *
> * SYNOPSIS
> */
> osm_signal_t
> -osm_mcast_mgr_process_mgrp_cb(IN void *const Context1, IN void *const Context2);
> +osm_mcast_mgr_process_mgroups(IN osm_mcast_mgr_t *p_mgr);
> /*
> * PARAMETERS
> -* (Context1) p_mgr
> +* p_mgr
> * [in] Pointer to an osm_mcast_mgr_t object.
> *
> -* (Context2) p_mgrp
> -* [in] Pointer to the multicast group to process.
> -*
> * RETURN VALUES
> * IB_SUCCESS
> *
> diff --git a/opensm/include/opensm/osm_multicast.h b/opensm/include/opensm/osm_multicast.h
> index 729a2ea..f442a45 100644
> --- a/opensm/include/opensm/osm_multicast.h
> +++ b/opensm/include/opensm/osm_multicast.h
> @@ -50,6 +50,7 @@
>
> #include <iba/ib_types.h>
> #include <complib/cl_qmap.h>
> +#include <complib/cl_qlist.h>
> #include <complib/cl_spinlock.h>
> #include <opensm/osm_base.h>
> #include <opensm/osm_mtree.h>
> @@ -121,6 +122,7 @@ const char *osm_get_mcast_req_type_str(IN osm_mcast_req_type_t req_type);
> * SYNOPSIS
> */
> typedef struct osm_mcast_mgr_ctxt {
> + cl_list_item_t list_item;
> ib_net16_t mlid;
> osm_mcast_req_type_t req_type;
> ib_net64_t port_guid;
> diff --git a/opensm/include/opensm/osm_sm.h b/opensm/include/opensm/osm_sm.h
> index 4c6ce27..a676cd6 100644
> --- a/opensm/include/opensm/osm_sm.h
> +++ b/opensm/include/opensm/osm_sm.h
> @@ -140,6 +140,8 @@ typedef struct osm_sm {
> cl_dispatcher_t *p_disp;
> cl_plock_t *p_lock;
> atomic32_t sm_trans_id;
> + cl_spinlock_t mgrp_lock;
> + cl_qlist_t mgrp_list;
> osm_req_t req;
> osm_resp_t resp;
> osm_ni_rcv_t ni_rcv;
> diff --git a/opensm/include/opensm/osm_state_mgr.h b/opensm/include/opensm/osm_state_mgr.h
> index dada097..f51593a 100644
> --- a/opensm/include/opensm/osm_state_mgr.h
> +++ b/opensm/include/opensm/osm_state_mgr.h
> @@ -109,8 +109,6 @@ typedef struct _osm_state_mgr {
> osm_stats_t *p_stats;
> struct _osm_sm_state_mgr *p_sm_state_mgr;
> const osm_sm_mad_ctrl_t *p_mad_ctrl;
> - cl_spinlock_t idle_lock;
> - cl_qlist_t idle_time_list;
> cl_plock_t *p_lock;
> cl_event_t *p_subnet_up_event;
> osm_sm_state_t state;
> @@ -172,99 +170,6 @@ typedef struct _osm_state_mgr {
> * State Manager object
> *********/
>
> -/****s* OpenSM: State Manager/_osm_idle_item
> -* NAME
> -* _osm_idle_item
> -*
> -* DESCRIPTION
> -* Idle item.
> -*
> -* SYNOPSIS
> -*/
> -
> -typedef osm_signal_t(*osm_pfn_start_t) (IN void *context1, IN void *context2);
> -
> -typedef void
> - (*osm_pfn_done_t) (IN void *context1, IN void *context2);
> -
> -typedef struct _osm_idle_item {
> - cl_list_item_t list_item;
> - void *context1;
> - void *context2;
> - osm_pfn_start_t pfn_start;
> - osm_pfn_done_t pfn_done;
> -} osm_idle_item_t;
> -
> -/*
> -* FIELDS
> -* list_item
> -* list item.
> -*
> -* context1
> -* Context pointer
> -*
> -* context2
> -* Context pointer
> -*
> -* pfn_start
> -* Pointer to the start function.
> -*
> -* pfn_done
> -* Pointer to the dine function.
> -* SEE ALSO
> -* State Manager object
> -*********/
> -
> -/****f* OpenSM: State Manager/osm_state_mgr_process_idle
> -* NAME
> -* osm_state_mgr_process_idle
> -*
> -* DESCRIPTION
> -* Formulates the osm_idle_item and inserts it into the queue and
> -* signals the state manager.
> -*
> -* SYNOPSIS
> -*/
> -
> -ib_api_status_t
> -osm_state_mgr_process_idle(IN osm_state_mgr_t * const p_mgr,
> - IN osm_pfn_start_t pfn_start,
> - IN osm_pfn_done_t pfn_done,
> - void *context1, void *context2);
> -
> -/*
> -* PARAMETERS
> -* p_mgr
> -* [in] Pointer to a State Manager object to construct.
> -*
> -* pfn_start
> -* [in] Pointer the start function which will be called at
> -* idle time.
> -*
> -* pfn_done
> -* [in] pointer the done function which will be called
> -* when outstanding smps is zero
> -*
> -* context1
> -* [in] Pointer to void
> -*
> -* context2
> -* [in] Pointer to void
> -*
> -* RETURN VALUE
> -* IB_SUCCESS or IB_ERROR
> -*
> -* NOTES
> -* Allows osm_state_mgr_destroy
> -*
> -* Calling osm_state_mgr_construct is a prerequisite to calling any other
> -* method except osm_state_mgr_init.
> -*
> -* SEE ALSO
> -* State Manager object, osm_state_mgr_init,
> -* osm_state_mgr_destroy
> -*********/
> -
> /****f* OpenSM: State Manager/osm_state_mgr_construct
> * NAME
> * osm_state_mgr_construct
> diff --git a/opensm/opensm/osm_mcast_mgr.c b/opensm/opensm/osm_mcast_mgr.c
> index 50b95fd..f51a45a 100644
> --- a/opensm/opensm/osm_mcast_mgr.c
> +++ b/opensm/opensm/osm_mcast_mgr.c
> @@ -815,7 +815,7 @@ static osm_mtree_node_t *__osm_mcast_mgr_branch(osm_mcast_mgr_t * const p_mgr,
> }
>
> free(list_array);
> - Exit:
> +Exit:
> OSM_LOG_EXIT(p_mgr->p_log);
> return (p_mtn);
> }
> @@ -932,7 +932,7 @@ __osm_mcast_mgr_build_spanning_tree(osm_mcast_mgr_t * const p_mgr,
> "Configured MLID 0x%X for %u ports, max tree depth = %u\n",
> cl_ntoh16(osm_mgrp_get_mlid(p_mgrp)), count, max_depth);
>
> - Exit:
> +Exit:
> OSM_LOG_EXIT(p_mgr->p_log);
> return (status);
> }
> @@ -1171,7 +1171,7 @@ osm_mcast_mgr_process_single(IN osm_mcast_mgr_t * const p_mgr,
> }
> }
>
> - Exit:
> +Exit:
> OSM_LOG_EXIT(p_mgr->p_log);
> return (status);
> }
> @@ -1254,63 +1254,55 @@ osm_mcast_mgr_process_tree(IN osm_mcast_mgr_t * const p_mgr,
> port_guid);
> }
>
> - Exit:
> +Exit:
> OSM_LOG_EXIT(p_mgr->p_log);
> return (status);
> }
>
> /**********************************************************************
> Process the entire group.
> -
> NOTE : The lock should be held externally!
> **********************************************************************/
> -static osm_signal_t
> -osm_mcast_mgr_process_mgrp(IN osm_mcast_mgr_t * const p_mgr,
> - IN osm_mgrp_t * const p_mgrp,
> - IN osm_mcast_req_type_t req_type,
> - IN ib_net64_t port_guid)
> +static ib_api_status_t
> +mcast_mgr_process_mgrp(IN osm_mcast_mgr_t * const p_mgr,
> + IN osm_mgrp_t * const p_mgrp,
> + IN osm_mcast_req_type_t req_type,
> + IN ib_net64_t port_guid)
> {
> - osm_signal_t signal = OSM_SIGNAL_DONE;
> ib_api_status_t status;
> - osm_switch_t *p_sw;
> - cl_qmap_t *p_sw_tbl;
> - boolean_t pending_transactions = FALSE;
>
> OSM_LOG_ENTER(p_mgr->p_log, osm_mcast_mgr_process_mgrp);
>
> - p_sw_tbl = &p_mgr->p_subn->sw_guid_tbl;
> -
> status = osm_mcast_mgr_process_tree(p_mgr, p_mgrp, req_type, port_guid);
> if (status != IB_SUCCESS) {
> osm_log(p_mgr->p_log, OSM_LOG_ERROR,
> - "osm_mcast_mgr_process_mgrp: ERR 0A19: "
> + "mcast_mgr_process_mgrp: ERR 0A19: "
> "Unable to create spanning tree (%s)\n",
> ib_get_err_str(status));
> -
> goto Exit;
> }
> + p_mgrp->last_tree_id = p_mgrp->last_change_id;
>
> - /*
> - Walk the switches and download the tables for each.
> + /* Remove MGRP only if osm_mcm_port_t count is 0 and
> + * Not a well known group
> */
> - p_sw = (osm_switch_t *) cl_qmap_head(p_sw_tbl);
> - while (p_sw != (osm_switch_t *) cl_qmap_end(p_sw_tbl)) {
> - signal = __osm_mcast_mgr_set_tbl(p_mgr, p_sw);
> - if (signal == OSM_SIGNAL_DONE_PENDING)
> - pending_transactions = TRUE;
> -
> - p_sw = (osm_switch_t *) cl_qmap_next(&p_sw->map_item);
> + if (cl_qmap_count(&p_mgrp->mcm_port_tbl) == 0 && !p_mgrp->well_known) {
> + osm_log(p_mgr->p_log, OSM_LOG_DEBUG,
> + "mcast_mgr_process_mgrp: "
> + "Destroying mgrp with lid:0x%X\n",
> + cl_ntoh16(p_mgrp->mlid));
> + /* Send a Report to any InformInfo registered for
> + Trap 67 : MCGroup delete */
> + osm_mgrp_send_delete_notice(p_mgr->p_subn, p_mgr->p_log,
> + p_mgrp);
> + cl_qmap_remove_item(&p_mgr->p_subn->mgrp_mlid_tbl,
> + (cl_map_item_t *) p_mgrp);
> + osm_mgrp_delete(p_mgrp);
> }
>
> - osm_dump_mcast_routes(p_mgr->p_subn->p_osm);
> -
> - Exit:
> +Exit:
> OSM_LOG_EXIT(p_mgr->p_log);
> -
> - if (pending_transactions == TRUE)
> - return (OSM_SIGNAL_DONE_PENDING);
> - else
> - return (OSM_SIGNAL_DONE);
> + return status;
> }
>
> /**********************************************************************
> @@ -1321,14 +1313,13 @@ osm_signal_t osm_mcast_mgr_process(IN osm_mcast_mgr_t * const p_mgr)
> osm_switch_t *p_sw;
> cl_qmap_t *p_sw_tbl;
> cl_qmap_t *p_mcast_tbl;
> + cl_qlist_t *p_list = &p_mgr->p_subn->p_osm->sm.mgrp_list;
> osm_mgrp_t *p_mgrp;
> - ib_api_status_t status;
> boolean_t pending_transactions = FALSE;
>
> OSM_LOG_ENTER(p_mgr->p_log, osm_mcast_mgr_process);
>
> p_sw_tbl = &p_mgr->p_subn->sw_guid_tbl;
> -
> p_mcast_tbl = &p_mgr->p_subn->mgrp_mlid_tbl;
> /*
> While holding the lock, iterate over all the established
> @@ -1343,16 +1334,8 @@ osm_signal_t osm_mcast_mgr_process(IN osm_mcast_mgr_t * const p_mgr)
> /* We reached here due to some change that caused a heavy sweep
> of the subnet. Not due to a specific multicast request.
> So the request type is subnet_change and the port guid is 0. */
> - status = osm_mcast_mgr_process_tree(p_mgr, p_mgrp,
> - OSM_MCAST_REQ_TYPE_SUBNET_CHANGE,
> - 0);
> - if (status != IB_SUCCESS) {
> - osm_log(p_mgr->p_log, OSM_LOG_ERROR,
> - "osm_mcast_mgr_process: ERR 0A20: "
> - "Unable to create spanning tree (%s)\n",
> - ib_get_err_str(status));
> - }
> -
> + mcast_mgr_process_mgrp(p_mgr, p_mgrp,
> + OSM_MCAST_REQ_TYPE_SUBNET_CHANGE, 0);
> p_mgrp = (osm_mgrp_t *) cl_qmap_next(&p_mgrp->map_item);
> }
>
> @@ -1364,10 +1347,14 @@ osm_signal_t osm_mcast_mgr_process(IN osm_mcast_mgr_t * const p_mgr)
> signal = __osm_mcast_mgr_set_tbl(p_mgr, p_sw);
> if (signal == OSM_SIGNAL_DONE_PENDING)
> pending_transactions = TRUE;
> -
> p_sw = (osm_switch_t *) cl_qmap_next(&p_sw->map_item);
> }
>
> + while (!cl_is_qlist_empty(p_list)) {
> + cl_list_item_t *p = cl_qlist_remove_head(p_list);
> + free(p);
> + }
> +
> CL_PLOCK_RELEASE(p_mgr->p_lock);
>
> OSM_LOG_EXIT(p_mgr->p_log);
> @@ -1395,79 +1382,79 @@ osm_mgrp_t *__get_mgrp_by_mlid(IN osm_mcast_mgr_t * const p_mgr,
>
> /**********************************************************************
> This is the function that is invoked during idle time to handle the
> - process request. Context1 is simply the osm_mcast_mgr_t*, Context2
> - hold the mlid, port guid and action (join/leave/delete) required.
> + process request for mcast groups where join/leave/delete was required.
> **********************************************************************/
> -osm_signal_t
> -osm_mcast_mgr_process_mgrp_cb(IN void *const Context1, IN void *const Context2)
> +osm_signal_t osm_mcast_mgr_process_mgroups(osm_mcast_mgr_t * p_mgr)
> {
> - osm_mcast_mgr_t *p_mgr = (osm_mcast_mgr_t *) Context1;
> + cl_qlist_t *p_list = &p_mgr->p_subn->p_osm->sm.mgrp_list;
> + osm_switch_t *p_sw;
> + cl_qmap_t *p_sw_tbl;
> osm_mgrp_t *p_mgrp;
> ib_net16_t mlid;
> - osm_signal_t signal = OSM_SIGNAL_DONE;
> - osm_mcast_mgr_ctxt_t *p_ctxt = (osm_mcast_mgr_ctxt_t *) Context2;
> - osm_mcast_req_type_t req_type = p_ctxt->req_type;
> - ib_net64_t port_guid = p_ctxt->port_guid;
> -
> - OSM_LOG_ENTER(p_mgr->p_log, osm_mcast_mgr_process_mgrp_cb);
> -
> - /* nice copy no warning on size diff */
> - memcpy(&mlid, &p_ctxt->mlid, sizeof(mlid));
> + osm_signal_t ret, signal = OSM_SIGNAL_DONE;
> + osm_mcast_mgr_ctxt_t *ctx;
> + osm_mcast_req_type_t req_type;
> + ib_net64_t port_guid;
>
> - /* we can destroy the context now */
> - free(p_ctxt);
> + OSM_LOG_ENTER(p_mgr->p_log, osm_mcast_mgr_process_mgroups);
>
> /* we need a lock to make sure the p_mgrp is not change other ways */
> CL_PLOCK_EXCL_ACQUIRE(p_mgr->p_lock);
> - p_mgrp = __get_mgrp_by_mlid(p_mgr, mlid);
>
> - /* since we delayed the execution we prefer to pass the
> - mlid as the mgrp identifier and then find it or abort */
> + if (cl_is_qlist_empty(p_list)) {
> + CL_PLOCK_RELEASE(p_mgr->p_lock);
> + return OSM_SIGNAL_NONE;
> + }
> +
> + while (!cl_is_qlist_empty(p_list)) {
> + ctx = (osm_mcast_mgr_ctxt_t *) cl_qlist_remove_head(p_list);
> + req_type = ctx->req_type;
> + port_guid = ctx->port_guid;
> +
> + /* nice copy no warning on size diff */
> + memcpy(&mlid, &ctx->mlid, sizeof(mlid));
>
> - if (p_mgrp) {
> + /* we can destroy the context now */
> + free(ctx);
> +
> + /* since we delayed the execution we prefer to pass the
> + mlid as the mgrp identifier and then find it or abort */
> + p_mgrp = __get_mgrp_by_mlid(p_mgr, mlid);
> + if (!p_mgrp)
> + continue;
>
> - /* if there was no change from the last time we processed the group
> - we can skip doing anything
> + /* if there was no change from the last time
> + * we processed the group we can skip doing anything
> */
> if (p_mgrp->last_change_id == p_mgrp->last_tree_id) {
> osm_log(p_mgr->p_log, OSM_LOG_DEBUG,
> - "osm_mcast_mgr_process_mgrp_cb: "
> + "osm_mcast_mgr_process_mgroups: "
> "Skip processing mgrp with lid:0x%X change id:%u\n",
> cl_ntoh16(mlid), p_mgrp->last_change_id);
> - } else {
> - osm_log(p_mgr->p_log, OSM_LOG_DEBUG,
> - "osm_mcast_mgr_process_mgrp_cb: "
> - "Processing mgrp with lid:0x%X change id:%u\n",
> - cl_ntoh16(mlid), p_mgrp->last_change_id);
> -
> - signal =
> - osm_mcast_mgr_process_mgrp(p_mgr, p_mgrp, req_type,
> - port_guid);
> - p_mgrp->last_tree_id = p_mgrp->last_change_id;
> + continue;
> }
>
> - /* Remove MGRP only if osm_mcm_port_t count is 0 and
> - * Not a well known group
> - */
> - if ((0x0 == cl_qmap_count(&p_mgrp->mcm_port_tbl)) &&
> - (p_mgrp->well_known == FALSE)) {
> - osm_log(p_mgr->p_log, OSM_LOG_DEBUG,
> - "osm_mcast_mgr_process_mgrp_cb: "
> - "Destroying mgrp with lid:0x%X\n",
> - cl_ntoh16(mlid));
> -
> - /* Send a Report to any InformInfo registered for
> - Trap 67 : MCGroup delete */
> - osm_mgrp_send_delete_notice(p_mgr->p_subn, p_mgr->p_log,
> - p_mgrp);
> -
> - cl_qmap_remove_item(&p_mgr->p_subn->mgrp_mlid_tbl,
> - (cl_map_item_t *) p_mgrp);
> + osm_log(p_mgr->p_log, OSM_LOG_DEBUG,
> + "osm_mcast_mgr_process_mgroups: "
> + "Processing mgrp with lid:0x%X change id:%u\n",
> + cl_ntoh16(mlid), p_mgrp->last_change_id);
> + mcast_mgr_process_mgrp(p_mgr, p_mgrp, req_type, port_guid);
> + }
>
> - osm_mgrp_delete(p_mgrp);
> - }
> + /*
> + Walk the switches and download the tables for each.
> + */
> + p_sw_tbl = &p_mgr->p_subn->sw_guid_tbl;
> + p_sw = (osm_switch_t *) cl_qmap_head(p_sw_tbl);
> + while (p_sw != (osm_switch_t *) cl_qmap_end(p_sw_tbl)) {
> + ret = __osm_mcast_mgr_set_tbl(p_mgr, p_sw);
> + if (ret == OSM_SIGNAL_DONE_PENDING)
> + signal = ret;
> + p_sw = (osm_switch_t *) cl_qmap_next(&p_sw->map_item);
> }
>
> + osm_dump_mcast_routes(p_mgr->p_subn->p_osm);
> +
> CL_PLOCK_RELEASE(p_mgr->p_lock);
> OSM_LOG_EXIT(p_mgr->p_log);
> return signal;
> diff --git a/opensm/opensm/osm_sm.c b/opensm/opensm/osm_sm.c
> index 88e6d4a..b295a77 100644
> --- a/opensm/opensm/osm_sm.c
> +++ b/opensm/opensm/osm_sm.c
> @@ -144,6 +144,7 @@ void osm_sm_construct(IN osm_sm_t * const p_sm)
> cl_event_construct(&p_sm->signal_event);
> cl_event_construct(&p_sm->subnet_up_event);
> cl_thread_construct(&p_sm->sweeper);
> + cl_spinlock_construct(&p_sm->mgrp_lock);
> osm_req_construct(&p_sm->req);
> osm_resp_construct(&p_sm->resp);
> osm_ni_rcv_construct(&p_sm->ni_rcv);
> @@ -245,6 +246,7 @@ void osm_sm_destroy(IN osm_sm_t * const p_sm)
> cl_event_destroy(&p_sm->signal_event);
> cl_event_destroy(&p_sm->subnet_up_event);
> cl_spinlock_destroy(&p_sm->signal_lock);
> + cl_spinlock_destroy(&p_sm->mgrp_lock);
>
> osm_log(p_sm->p_log, OSM_LOG_SYS, "Exiting SM\n"); /* Format Waived */
> OSM_LOG_EXIT(p_sm->p_log);
> @@ -292,6 +294,12 @@ osm_sm_init(IN osm_sm_t * const p_sm,
> if (status != CL_SUCCESS)
> goto Exit;
>
> + cl_qlist_init(&p_sm->mgrp_list);
> +
> + status = cl_spinlock_init(&p_sm->mgrp_lock);
> + if (status != CL_SUCCESS)
> + goto Exit;
> +
> status = osm_sm_mad_ctrl_init(&p_sm->mad_ctrl,
> p_sm->p_subn,
> p_sm->p_mad_pool,
> @@ -551,32 +559,43 @@ osm_sm_bind(IN osm_sm_t * const p_sm, IN const ib_net64_t port_guid)
> /**********************************************************************
> **********************************************************************/
> static ib_api_status_t
> -__osm_sm_mgrp_connect(IN osm_sm_t * const p_sm,
> +__osm_sm_mgrp_process(IN osm_sm_t * const p_sm,
> IN osm_mgrp_t * const p_mgrp,
> IN const ib_net64_t port_guid,
> IN osm_mcast_req_type_t req_type)
> {
> - ib_api_status_t status;
> osm_mcast_mgr_ctxt_t *ctx2;
>
> - OSM_LOG_ENTER(p_sm->p_log, __osm_sm_mgrp_connect);
> -
> /*
> * 'Schedule' all the QP0 traffic for when the state manager
> * isn't busy trying to do something else.
> */
> ctx2 = (osm_mcast_mgr_ctxt_t *) malloc(sizeof(osm_mcast_mgr_ctxt_t));
> + if (!ctx2)
> + return IB_ERROR;
> + memset(ctx2, 0, sizeof(*ctx2));
> memcpy(&ctx2->mlid, &p_mgrp->mlid, sizeof(p_mgrp->mlid));
> ctx2->req_type = req_type;
> ctx2->port_guid = port_guid;
>
> - status = osm_state_mgr_process_idle(&p_sm->state_mgr,
> - osm_mcast_mgr_process_mgrp_cb,
> - NULL, &p_sm->mcast_mgr,
> - (void *)ctx2);
> + cl_spinlock_acquire(&p_sm->mgrp_lock);
> + cl_qlist_insert_tail(&p_sm->mgrp_list, &ctx2->list_item);
> + cl_spinlock_release(&p_sm->mgrp_lock);
>
> - OSM_LOG_EXIT(p_sm->p_log);
> - return (status);
> + osm_sm_signal(p_sm, OSM_SIGNAL_IDLE_TIME_PROCESS_REQUEST);
> +
> + return IB_SUCCESS;
> +}
> +
> +/**********************************************************************
> + **********************************************************************/
> +static ib_api_status_t
> +__osm_sm_mgrp_connect(IN osm_sm_t * const p_sm,
> + IN osm_mgrp_t * const p_mgrp,
> + IN const ib_net64_t port_guid,
> + IN osm_mcast_req_type_t req_type)
> +{
> + return __osm_sm_mgrp_process(p_sm, p_mgrp, port_guid, req_type);
> }
>
> /**********************************************************************
> @@ -586,31 +605,7 @@ __osm_sm_mgrp_disconnect(IN osm_sm_t * const p_sm,
> IN osm_mgrp_t * const p_mgrp,
> IN const ib_net64_t port_guid)
> {
> - ib_api_status_t status;
> - osm_mcast_mgr_ctxt_t *ctx2;
> -
> - OSM_LOG_ENTER(p_sm->p_log, __osm_sm_mgrp_disconnect);
> -
> - /*
> - * 'Schedule' all the QP0 traffic for when the state manager
> - * isn't busy trying to do something else.
> - */
> - ctx2 = (osm_mcast_mgr_ctxt_t *) malloc(sizeof(osm_mcast_mgr_ctxt_t));
> - memcpy(&ctx2->mlid, &p_mgrp->mlid, sizeof(p_mgrp->mlid));
> - ctx2->req_type = OSM_MCAST_REQ_TYPE_LEAVE;
> - ctx2->port_guid = port_guid;
> -
> - status = osm_state_mgr_process_idle(&p_sm->state_mgr,
> - osm_mcast_mgr_process_mgrp_cb,
> - NULL, &p_sm->mcast_mgr, ctx2);
> - if (status != IB_SUCCESS) {
> - osm_log(p_sm->p_log, OSM_LOG_ERROR,
> - "__osm_sm_mgrp_disconnect: ERR 2E11: "
> - "Failure processing multicast group (%s)\n",
> - ib_get_err_str(status));
> - }
> -
> - OSM_LOG_EXIT(p_sm->p_log);
> + __osm_sm_mgrp_process(p_sm, p_mgrp, port_guid, OSM_MCAST_REQ_TYPE_LEAVE);
> }
>
> /**********************************************************************
> @@ -719,8 +714,8 @@ osm_sm_mcgrp_join(IN osm_sm_t * const p_sm,
> goto Exit;
> }
>
> - CL_PLOCK_RELEASE(p_sm->p_lock);
> status = __osm_sm_mgrp_connect(p_sm, p_mgrp, port_guid, req_type);
> + CL_PLOCK_RELEASE(p_sm->p_lock);
>
> Exit:
> OSM_LOG_EXIT(p_sm->p_log);
> @@ -782,9 +777,8 @@ osm_sm_mcgrp_leave(IN osm_sm_t * const p_sm,
>
> osm_port_remove_mgrp(p_port, mlid);
>
> - CL_PLOCK_RELEASE(p_sm->p_lock);
> -
> __osm_sm_mgrp_disconnect(p_sm, p_mgrp, port_guid);
> + CL_PLOCK_RELEASE(p_sm->p_lock);
>
> Exit:
> OSM_LOG_EXIT(p_sm->p_log);
> diff --git a/opensm/opensm/osm_state_mgr.c b/opensm/opensm/osm_state_mgr.c
> index 5c39f11..d4dd782 100644
> --- a/opensm/opensm/osm_state_mgr.c
> +++ b/opensm/opensm/osm_state_mgr.c
> @@ -76,7 +76,6 @@ osm_signal_t osm_qos_setup(IN osm_opensm_t * p_osm);
> void osm_state_mgr_construct(IN osm_state_mgr_t * const p_mgr)
> {
> memset(p_mgr, 0, sizeof(*p_mgr));
> - cl_spinlock_construct(&p_mgr->idle_lock);
> p_mgr->state = OSM_SM_STATE_INIT;
> }
>
> @@ -88,9 +87,6 @@ void osm_state_mgr_destroy(IN osm_state_mgr_t * const p_mgr)
>
> OSM_LOG_ENTER(p_mgr->p_log, osm_state_mgr_destroy);
>
> - /* destroy the locks */
> - cl_spinlock_destroy(&p_mgr->idle_lock);
> -
> OSM_LOG_EXIT(p_mgr->p_log);
> }
>
> @@ -112,8 +108,6 @@ osm_state_mgr_init(IN osm_state_mgr_t * const p_mgr,
> IN cl_event_t * const p_subnet_up_event,
> IN osm_log_t * const p_log)
> {
> - cl_status_t status;
> -
> OSM_LOG_ENTER(p_log, osm_state_mgr_init);
>
> CL_ASSERT(p_subn);
> @@ -145,17 +139,8 @@ osm_state_mgr_init(IN osm_state_mgr_t * const p_mgr,
> p_mgr->p_lock = p_lock;
> p_mgr->p_subnet_up_event = p_subnet_up_event;
>
> - cl_qlist_init(&p_mgr->idle_time_list);
> -
> - status = cl_spinlock_init(&p_mgr->idle_lock);
> - if (status != CL_SUCCESS) {
> - osm_log(p_mgr->p_log, OSM_LOG_ERROR,
> - "osm_state_mgr_init: ERR 3302: "
> - "Spinlock init failed (%s)\n", CL_STATUS_MSG(status));
> - }
> -
> OSM_LOG_EXIT(p_mgr->p_log);
> - return (status);
> + return IB_SUCCESS;
> }
>
> /**********************************************************************
> @@ -989,79 +974,6 @@ static ib_api_status_t __osm_state_mgr_light_sweep_start(IN osm_state_mgr_t *
> }
>
> /**********************************************************************
> - **********************************************************************/
> -static void __process_idle_time_queue_done(IN osm_state_mgr_t * const p_mgr)
> -{
> - cl_qlist_t *p_list = &p_mgr->idle_time_list;
> - cl_list_item_t *p_list_item;
> - osm_idle_item_t *p_process_item;
> -
> - OSM_LOG_ENTER(p_mgr->p_log, __process_idle_time_queue_done);
> -
> - cl_spinlock_acquire(&p_mgr->idle_lock);
> - p_list_item = cl_qlist_remove_head(p_list);
> -
> - if (p_list_item == cl_qlist_end(p_list)) {
> - cl_spinlock_release(&p_mgr->idle_lock);
> - osm_log(p_mgr->p_log, OSM_LOG_ERROR,
> - "__process_idle_time_queue_done: ERR 3314: "
> - "Idle time queue is empty\n");
> - return;
> - }
> - cl_spinlock_release(&p_mgr->idle_lock);
> -
> - p_process_item = (osm_idle_item_t *) p_list_item;
> -
> - if (p_process_item->pfn_done) {
> -
> - p_process_item->pfn_done(p_process_item->context1,
> - p_process_item->context2);
> - }
> -
> - free(p_process_item);
> -
> - OSM_LOG_EXIT(p_mgr->p_log);
> - return;
> -}
> -
> -/**********************************************************************
> - **********************************************************************/
> -static osm_signal_t __process_idle_time_queue_start(IN osm_state_mgr_t *
> - const p_mgr)
> -{
> - cl_qlist_t *p_list = &p_mgr->idle_time_list;
> - cl_list_item_t *p_list_item;
> - osm_idle_item_t *p_process_item;
> - osm_signal_t signal;
> -
> - OSM_LOG_ENTER(p_mgr->p_log, __process_idle_time_queue_start);
> -
> - cl_spinlock_acquire(&p_mgr->idle_lock);
> -
> - p_list_item = cl_qlist_head(p_list);
> - if (p_list_item == cl_qlist_end(p_list)) {
> - cl_spinlock_release(&p_mgr->idle_lock);
> - OSM_LOG_EXIT(p_mgr->p_log);
> - return OSM_SIGNAL_NONE;
> - }
> -
> - cl_spinlock_release(&p_mgr->idle_lock);
> -
> - p_process_item = (osm_idle_item_t *) p_list_item;
> -
> - CL_ASSERT(p_process_item->pfn_start);
> -
> - signal =
> - p_process_item->pfn_start(p_process_item->context1,
> - p_process_item->context2);
> -
> - CL_ASSERT(signal != OSM_SIGNAL_NONE);
> -
> - OSM_LOG_EXIT(p_mgr->p_log);
> - return signal;
> -}
> -
> -/**********************************************************************
> * Go over all the remote SMs (as updated in the sm_guid_tbl).
> * Find if there is a remote sm that is a master SM.
> * If there is a remote master SM - return a pointer to it,
> @@ -1558,7 +1470,7 @@ void osm_state_mgr_process(IN osm_state_mgr_t * const p_mgr,
> case OSM_SM_STATE_PROCESS_REQUEST:
> switch (signal) {
> case OSM_SIGNAL_IDLE_TIME_PROCESS:
> - signal = __process_idle_time_queue_start(p_mgr);
> + signal = osm_mcast_mgr_process_mgroups(p_mgr->p_mcast_mgr);
> switch (signal) {
> case OSM_SIGNAL_NONE:
> p_mgr->state = OSM_SM_STATE_IDLE;
> @@ -1604,14 +1516,6 @@ void osm_state_mgr_process(IN osm_state_mgr_t * const p_mgr,
> switch (signal) {
> case OSM_SIGNAL_NO_PENDING_TRANSACTIONS:
> case OSM_SIGNAL_DONE:
> - /* CALL the done function */
> - __process_idle_time_queue_done(p_mgr);
> -
> - /*
> - * Set the signal to OSM_SIGNAL_IDLE_TIME_PROCESS
> - * so that the next element in the queue gets processed
> - */
> -
> signal = OSM_SIGNAL_IDLE_TIME_PROCESS;
> p_mgr->state = OSM_SM_STATE_PROCESS_REQUEST;
> break;
> @@ -2424,41 +2328,3 @@ void osm_state_mgr_process(IN osm_state_mgr_t * const p_mgr,
>
> OSM_LOG_EXIT(p_mgr->p_log);
> }
> -
> -/**********************************************************************
> - **********************************************************************/
> -ib_api_status_t
> -osm_state_mgr_process_idle(IN osm_state_mgr_t * const p_mgr,
> - IN osm_pfn_start_t pfn_start,
> - IN osm_pfn_done_t pfn_done, void *context1,
> - void *context2)
> -{
> - osm_idle_item_t *p_idle_item;
> -
> - OSM_LOG_ENTER(p_mgr->p_log, osm_state_mgr_process_idle);
> -
> - p_idle_item = malloc(sizeof(osm_idle_item_t));
> - if (p_idle_item == NULL) {
> - osm_log(p_mgr->p_log, OSM_LOG_ERROR,
> - "osm_state_mgr_process_idle: ERR 3321: "
> - "insufficient memory\n");
> - return IB_ERROR;
> - }
> -
> - memset(p_idle_item, 0, sizeof(osm_idle_item_t));
> - p_idle_item->pfn_start = pfn_start;
> - p_idle_item->pfn_done = pfn_done;
> - p_idle_item->context1 = context1;
> - p_idle_item->context2 = context2;
> -
> - cl_spinlock_acquire(&p_mgr->idle_lock);
> - cl_qlist_insert_tail(&p_mgr->idle_time_list, &p_idle_item->list_item);
> - cl_spinlock_release(&p_mgr->idle_lock);
> -
> - osm_sm_signal(&p_mgr->p_subn->p_osm->sm,
> - OSM_SIGNAL_IDLE_TIME_PROCESS_REQUEST);
> -
> - OSM_LOG_EXIT(p_mgr->p_log);
> -
> - return IB_SUCCESS;
> -}
More information about the general
mailing list