[mynewt-nimble] 02/04: nimble/ll: Optimize sched restart

[an...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

andk pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/mynewt-nimble.git

commit 7f63bd094e0ae5a3596b36fc9a116c1bf52330fb
Author: Andrzej Kaczmarek <an...@codecoup.pl>
AuthorDate: Wed Oct 6 09:58:16 2021 +0200

    nimble/ll: Optimize sched restart
    
    When sched_q head changed we only should stop sched timer to make sure
    it does not fire after we leave critical section and we can restart it
    then without need for additional locking.
    
    We do not need to change rfmgmt because in case new sched_q head is
    later then previous one, rfmgmt will fire too early and we'll either
    disable it immediately on restart or keep it enabled for upcoming item,
    depending on timing.
---
 nimble/controller/src/ble_ll_sched.c | 31 ++++++++++++++++++++-----------
 1 file changed, 20 insertions(+), 11 deletions(-)

diff --git a/nimble/controller/src/ble_ll_sched.c b/nimble/controller/src/ble_ll_sched.c
index 03dea23..7a2e6f3 100644
--- a/nimble/controller/src/ble_ll_sched.c
+++ b/nimble/controller/src/ble_ll_sched.c
@@ -37,7 +37,6 @@
 #define BLE_LL_SCHED_MAX_DELAY_ANY      (0x7fffffff)
 
 static struct hal_timer g_ble_ll_sched_timer;
-static uint8_t g_ble_ll_sched_timer_running;
 
 uint8_t g_ble_ll_sched_offset_ticks;
 
@@ -63,6 +62,7 @@ typedef int (* ble_ll_sched_preempt_cb_t)(struct ble_ll_sched_item *sch,
 
 /* Queue for timers */
 static TAILQ_HEAD(ll_sched_qhead, ble_ll_sched_item) g_ble_ll_sched_q;
+static uint8_t g_ble_ll_sched_q_head_changed;
 
 #if MYNEWT_VAL(BLE_LL_STRICT_CONN_SCHEDULING)
 struct ble_ll_sched_obj g_ble_ll_sched_data;
@@ -155,13 +155,13 @@ ble_ll_sched_preempt(struct ble_ll_sched_item *sch,
 static inline void
 ble_ll_sched_q_head_changed(void)
 {
-    struct ble_ll_sched_item *first;
+    if (g_ble_ll_sched_q_head_changed) {
+        return;
+    }
 
-    g_ble_ll_sched_timer_running = 0;
-    os_cputime_timer_stop(&g_ble_ll_sched_timer);
+    g_ble_ll_sched_q_head_changed = 1;
 
-    first = TAILQ_FIRST(&g_ble_ll_sched_q);
-    ble_ll_rfmgmt_sched_changed(first);
+    os_cputime_timer_stop(&g_ble_ll_sched_timer);
 }
 
 static inline void
@@ -169,10 +169,17 @@ ble_ll_sched_restart(void)
 {
     struct ble_ll_sched_item *first;
 
+    if (!g_ble_ll_sched_q_head_changed) {
+        return;
+    }
+
+    g_ble_ll_sched_q_head_changed = 0;
+
     first = TAILQ_FIRST(&g_ble_ll_sched_q);
 
-    if (!g_ble_ll_sched_timer_running && first) {
-        g_ble_ll_sched_timer_running = 1;
+    ble_ll_rfmgmt_sched_changed(first);
+
+    if (first) {
         os_cputime_timer_start(&g_ble_ll_sched_timer, first->start_time);
     }
 }
@@ -1182,8 +1189,6 @@ ble_ll_sched_run(void *arg)
 
     BLE_LL_DEBUG_GPIO(SCHED_RUN, 1);
 
-    g_ble_ll_sched_timer_running = 0;
-
     /* Look through schedule queue */
     sch = TAILQ_FIRST(&g_ble_ll_sched_q);
     if (sch) {
@@ -1203,7 +1208,10 @@ ble_ll_sched_run(void *arg)
         /* Remove schedule item and execute the callback */
         TAILQ_REMOVE(&g_ble_ll_sched_q, sch, link);
         sch->enqueued = 0;
+        g_ble_ll_sched_q_head_changed = 1;
+
         ble_ll_sched_execute_item(sch);
+
         ble_ll_sched_restart();
     }
 
@@ -1295,7 +1303,6 @@ int ble_ll_sched_dtm(struct ble_ll_sched_item *sch)
 void
 ble_ll_sched_stop(void)
 {
-    g_ble_ll_sched_timer_running = 0;
     os_cputime_timer_stop(&g_ble_ll_sched_timer);
 }
 
@@ -1338,5 +1345,7 @@ ble_ll_sched_init(void)
         g_ble_ll_sched_data.sch_ticks_per_period;
 #endif
 
+    g_ble_ll_sched_q_head_changed = 0;
+
     return 0;
 }