event/octeontx: support worker enqueue
authorJerin Jacob <jerin.jacob@caviumnetworks.com>
Fri, 3 Mar 2017 17:28:00 +0000 (22:58 +0530)
committerJerin Jacob <jerin.jacob@caviumnetworks.com>
Tue, 4 Apr 2017 17:19:53 +0000 (19:19 +0200)
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Gage Eads <gage.eads@intel.com>
drivers/event/octeontx/ssovf_evdev.h
drivers/event/octeontx/ssovf_worker.c

index 16846b8..fac42f5 100644 (file)
@@ -187,4 +187,8 @@ ssovf_pmd_priv(const struct rte_eventdev *eventdev)
        return eventdev->data->dev_private;
 }
 
+uint16_t ssows_enq(void *port, const struct rte_event *ev);
+uint16_t ssows_enq_burst(void *port,
+               const struct rte_event ev[], uint16_t nb_events);
+
 #endif /* __SSOVF_EVDEV_H__ */
index 14f3426..050a0e6 100644 (file)
  */
 
 #include "ssovf_worker.h"
+
+static force_inline void
+ssows_new_event(struct ssows *ws, const struct rte_event *ev)
+{
+       const uint64_t event_ptr = ev->u64;
+       const uint32_t tag = (uint32_t)ev->event;
+       const uint8_t new_tt = ev->sched_type;
+       const uint8_t grp = ev->queue_id;
+
+       ssows_add_work(ws, event_ptr, tag, new_tt, grp);
+}
+
+static force_inline void
+ssows_fwd_swtag(struct ssows *ws, const struct rte_event *ev, const uint8_t grp)
+{
+       const uint8_t cur_tt = ws->cur_tt;
+       const uint8_t new_tt = ev->sched_type;
+       const uint32_t tag = (uint32_t)ev->event;
+       /*
+        * cur_tt/new_tt     SSO_SYNC_ORDERED SSO_SYNC_ATOMIC SSO_SYNC_UNTAGGED
+        *
+        * SSO_SYNC_ORDERED        norm           norm             untag
+        * SSO_SYNC_ATOMIC         norm           norm             untag
+        * SSO_SYNC_UNTAGGED       full           full             NOOP
+        */
+       if (unlikely(cur_tt == SSO_SYNC_UNTAGGED)) {
+               if (new_tt != SSO_SYNC_UNTAGGED) {
+                       ssows_swtag_full(ws, ev->u64, tag,
+                               new_tt, grp);
+               }
+       } else {
+               if (likely(new_tt != SSO_SYNC_UNTAGGED))
+                       ssows_swtag_norm(ws, tag, new_tt);
+               else
+                       ssows_swtag_untag(ws);
+       }
+       ws->swtag_req = 1;
+}
+
+#define OCT_EVENT_TYPE_GRP_FWD (RTE_EVENT_TYPE_MAX - 1)
+
+static force_inline void
+ssows_fwd_group(struct ssows *ws, const struct rte_event *ev, const uint8_t grp)
+{
+       const uint64_t event_ptr = ev->u64;
+       const uint32_t tag = (uint32_t)ev->event;
+       const uint8_t cur_tt = ws->cur_tt;
+       const uint8_t new_tt = ev->sched_type;
+
+       if (cur_tt == SSO_SYNC_ORDERED) {
+               /* Create unique tag based on custom event type and new grp */
+               uint32_t newtag = OCT_EVENT_TYPE_GRP_FWD << 28;
+
+               newtag |= grp << 20;
+               newtag |= tag;
+               ssows_swtag_norm(ws, newtag, SSO_SYNC_ATOMIC);
+               rte_smp_wmb();
+               ssows_swtag_wait(ws);
+       } else {
+               rte_smp_wmb();
+       }
+       ssows_add_work(ws, event_ptr, tag, new_tt, grp);
+}
+
+static force_inline void
+ssows_forward_event(struct ssows *ws, const struct rte_event *ev)
+{
+       const uint8_t grp = ev->queue_id;
+
+       /* Group hasn't changed, Use SWTAG to forward the event */
+       if (ws->cur_grp == grp)
+               ssows_fwd_swtag(ws, ev, grp);
+       else
+       /*
+        * Group has been changed for group based work pipelining,
+        * Use deschedule/add_work operation to transfer the event to
+        * new group/core
+        */
+               ssows_fwd_group(ws, ev, grp);
+}
+
+static force_inline void
+ssows_release_event(struct ssows *ws)
+{
+       if (likely(ws->cur_tt != SSO_SYNC_UNTAGGED))
+               ssows_swtag_untag(ws);
+}
+
+force_inline uint16_t __hot
+ssows_enq(void *port, const struct rte_event *ev)
+{
+       struct ssows *ws = port;
+       uint16_t ret = 1;
+
+       switch (ev->op) {
+       case RTE_EVENT_OP_NEW:
+               ssows_new_event(ws, ev);
+               break;
+       case RTE_EVENT_OP_FORWARD:
+               ssows_forward_event(ws, ev);
+               break;
+       case RTE_EVENT_OP_RELEASE:
+               ssows_release_event(ws);
+               break;
+       default:
+               ret = 0;
+       }
+       return ret;
+}
+
+uint16_t __hot
+ssows_enq_burst(void *port, const struct rte_event ev[], uint16_t nb_events)
+{
+       RTE_SET_USED(nb_events);
+       return ssows_enq(port, ev);
+}