15 files changed, 226 insertions, 269 deletions

diff --git a/include/litmus/event_group.h b/include/litmus/event_group.h
index b2a6a3ff5627..7b15a7e0412d 100644
--- a/include/litmus/event_group.h
+++ b/include/litmus/event_group.h
@@ -25,6 +25,7 @@ struct event_list {
         /* For timer firing */
         lt_t                            fire_time;
         struct hrtimer                  timer;
+        struct hrtimer_start_on_info    info;
 
         struct list_head    queue_node;  /* For event_queue */
         struct event_group* group; /* For callback    */
diff --git a/include/litmus/preempt.h b/include/litmus/preempt.h
index cbf315aa01e9..beb4d480f21b 100644
--- a/include/litmus/preempt.h
+++ b/include/litmus/preempt.h
@@ -13,7 +13,7 @@ extern DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, resched_state);
 
 #ifdef CONFIG_PREEMPT_STATE_TRACE
 const char* sched_state_name(int s);
-#define TRACE_STATE(fmt, args...) TRACE("SCHED_STATE " fmt, args)
+#define TRACE_STATE(fmt, args...) STRACE("SCHED_STATE " fmt, args)
 #else
 #define TRACE_STATE(fmt, args...) /* ignore */
 #endif
diff --git a/include/litmus/rt_param.h b/include/litmus/rt_param.h
index a8fe95b32c06..52a0116f7282 100644
--- a/include/litmus/rt_param.h
+++ b/include/litmus/rt_param.h
@@ -268,19 +268,14 @@ struct rt_param {
         /* Pointer to the page shared between userspace and kernel. */
         struct control_page * ctrl_page;
 
-        lt_t            total_tardy;
-        lt_t            max_tardy;
-        unsigned int    missed;
-
+#ifdef CONFIG_SCHED_TASK_TRACE
         lt_t load;
         lt_t flush;
-        int load_work;
-        int flush_work;
+#endif
 
-        lt_t max_exec_time;
-        lt_t tot_exec_time;
         lt_t last_exec_time;
         lt_t orig_cost;
+
         struct color_ctrl_page color_ctrl_page;
         struct dgl_group_req *req;
         enum server_state state;
diff --git a/include/litmus/sched_mc.h b/include/litmus/sched_mc.h
index 70f57cfd2706..567603c5ffff 100644
--- a/include/litmus/sched_mc.h
+++ b/include/litmus/sched_mc.h
@@ -55,6 +55,7 @@ struct ce_dom_data {
         struct rt_event event;
 #else
         struct hrtimer timer;
+        struct hrtimer_start_on_info timer_info;
 #endif
 };
 
diff --git a/include/litmus/sched_trace.h b/include/litmus/sched_trace.h
index 0580340d0170..40187826ef19 100644
--- a/include/litmus/sched_trace.h
+++ b/include/litmus/sched_trace.h
@@ -51,18 +51,9 @@ struct st_switch_away_data {	/* A process was switched away from on a given CPU.
         u64     exec_time;
 };
 
-/* changed: like everything */
 struct st_completion_data {     /* A job completed. */
-        u32     exec;
-        u16     flush_work;
-        u16     load_work;
-        u32     flush;
-        u32     load;
-        /* u8   forced:1;       /\* Set to 1 if job overran and kernel advanced to the */
-        /*                       * next task automatically; set to 0 otherwise. */
-        /*                       *\/ */
-        /* u8   __uflags:7; */
-        /* u8   __unused[7]; */
+        u64     when;
+        u64     load;
 };
 
 struct st_block_data {          /* A task blocks. */
@@ -86,19 +77,6 @@ struct st_sys_release_data {
         u64     release;
 };
 
-/* changed: easy enough to remove */
-struct st_task_exit_data {
-        u64     avg_exec_time;
-        u64     max_exec_time;
-};
-
-/* changed: calculate yoself */
-struct st_task_tardy_data {
-        u64     total_tardy;
-        u32     max_tardy;
-        u32     missed;
-};
-
 #define DATA(x) struct st_ ## x ## _data x;
 
 typedef enum {
@@ -113,9 +91,7 @@ typedef enum {
         ST_BLOCK,
         ST_RESUME,
         ST_ACTION,
-        ST_SYS_RELEASE,
-        ST_TASK_EXIT,
-        ST_TASK_TARDY,
+        ST_SYS_RELEASE
 } st_event_record_type_t;
 
 struct st_event_record {
@@ -134,8 +110,6 @@ struct st_event_record {
                 DATA(resume);
                 DATA(action);
                 DATA(sys_release);
-                DATA(task_exit);
-                DATA(task_tardy);
         } data;
 };
 
@@ -177,11 +151,6 @@ feather_callback void do_sched_trace_action(unsigned long id,
                                             unsigned long action);
 feather_callback void do_sched_trace_sys_release(unsigned long id,
                                                  lt_t* start);
-feather_callback void do_sched_trace_task_exit(unsigned long id,
-                                                     struct task_struct* task);
-feather_callback void do_sched_trace_task_tardy(unsigned long id,
-                                                     struct task_struct* task);
-
 #endif
 
 #else
@@ -306,20 +275,6 @@ feather_callback void do_sched_trace_task_tardy(unsigned long id,
                 trace_litmus_sys_release(when);                         \
         } while (0)
 
-#define sched_trace_task_exit(t)                                        \
-        do {                                                            \
-                SCHED_TRACE(SCHED_TRACE_BASE_ID + 11,                   \
-                        do_sched_trace_task_exit, t);                   \
-                trace_litmus_task_exit(t);                              \
-        } while (0)
-
-
-#define sched_trace_task_tardy(t)                                       \
-        do {                                                            \
-                SCHED_TRACE(SCHED_TRACE_BASE_ID + 12,                   \
-                        do_sched_trace_task_tardy, t);                  \
-        } while (0)
-
 #define QT_START lt_t _qt_start = litmus_clock()
 #define QT_END \
         sched_trace_log_message("%d P%d      [%s@%s:%d]: Took %llu\n\n", \
diff --git a/include/trace/events/litmus.h b/include/trace/events/litmus.h
index 95aae9460cbc..c2dbdc34eb42 100644
--- a/include/trace/events/litmus.h
+++ b/include/trace/events/litmus.h
@@ -278,28 +278,6 @@ TRACE_EVENT(litmus_sys_release,
 );
 
 /*
- * Trace task exit
- */
-TRACE_EVENT(litmus_task_exit,
-
-        TP_PROTO(struct task_struct *t),
-
-        TP_ARGS(t),
-
-        TP_STRUCT__entry(
-                __field( pid_t, pid )
-                __field( unsigned long long, max_exec_time )
-        ),
-
-        TP_fast_assign(
-                __entry->pid = t ? t->pid : 0;
-                __entry->max_exec_time = t ? t->rt_param.max_exec_time : 0;
-        ),
-
-        TP_printk("(%u) exit\n", __entry->pid)
-);
-
-/*
  * Containers
  */
 TRACE_EVENT(litmus_container_param,
diff --git a/litmus/ce_domain.c b/litmus/ce_domain.c
index ca7e6ae67cf3..c750b95581b9 100644
--- a/litmus/ce_domain.c
+++ b/litmus/ce_domain.c
@@ -93,6 +93,7 @@ void ce_domain_init(domain_t *dom,
         init_event(&dom_data->event, CRIT_LEVEL_A, ce_timer_callback,
                         event_list_alloc(GFP_ATOMIC));
 #else
+        hrtimer_start_on_info_init(&dom_data->timer_info);
         hrtimer_init(&dom_data->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
         dom_data->timer.function = ce_timer_callback;
 #endif
diff --git a/litmus/jobs.c b/litmus/jobs.c
index 097d7dd94d12..9e6de1b08982 100644
--- a/litmus/jobs.c
+++ b/litmus/jobs.c
@@ -116,25 +116,6 @@ void release_at(struct task_struct *t, lt_t start)
  */
 long complete_job(void)
 {
-        lt_t amount;
-        lt_t now = litmus_clock();
-        lt_t exec_time = tsk_rt(current)->job_params.exec_time;
-
-        /* Task statistic summaries */
-        tsk_rt(current)->tot_exec_time += exec_time;
-        if (lt_before(tsk_rt(current)->max_exec_time, exec_time))
-                tsk_rt(current)->max_exec_time = exec_time;
-
-        if (is_tardy(current, now)) {
-                TRACE_TASK(current, "is tardy, now: %llu, deadline: %llu\n",
-                           now, get_deadline(current));
-                amount = now - get_deadline(current);
-                if (lt_after(amount, tsk_rt(current)->max_tardy))
-                        tsk_rt(current)->max_tardy = amount;
-                tsk_rt(current)->total_tardy += amount;
-                ++tsk_rt(current)->missed;
-        }
-
         TRACE_TASK(current, "user complete\n");
 
         /* Mark that we do not execute anymore */
diff --git a/litmus/litmus.c b/litmus/litmus.c
index 3de9252b3223..6dd631cfda4d 100644
--- a/litmus/litmus.c
+++ b/litmus/litmus.c
@@ -490,9 +490,6 @@ void litmus_exit_task(struct task_struct* tsk)
 {
         if (is_realtime(tsk)) {
                 sched_trace_task_completion(tsk, 1);
-                sched_trace_task_exit(tsk);
-                sched_trace_task_tardy(tsk);
-
                 litmus->task_exit(tsk);
 
                 BUG_ON(bheap_node_in_heap(tsk_rt(tsk)->heap_node));
diff --git a/litmus/lockdown.c b/litmus/lockdown.c
index 09712554c5b9..bc946f7464c0 100644
--- a/litmus/lockdown.c
+++ b/litmus/lockdown.c
@@ -63,6 +63,14 @@ u32 color_read_in_mem(u32 lock_val, u32 unlock_val, void *start, void *end)
         return 0;
 }
 
+u32 color_read_in_mem_lock(u32 lock_val, u32 unlock_val, void *start, void *end)
+{
+        TRACE_CUR("Dummy read_in_mem: lock_val: 0x%x  unlock_val: 0x%x  "
+                        "start: 0x%p  end: 0x%p\n", lock_val, unlock_val,
+                        start, end);
+        return 0;
+}
+
 void set_lockdown(u32 lockdown_state)
 {
         TRACE_CUR("Dummy set_lockdown function lockdown_state: 0x%x\n",
diff --git a/litmus/preempt.c b/litmus/preempt.c
index 8f1304afea26..26c6c7a929d9 100644
--- a/litmus/preempt.c
+++ b/litmus/preempt.c
@@ -22,9 +22,10 @@ void sched_state_will_schedule(struct task_struct* tsk)
          */
         if (likely(task_cpu(tsk) == smp_processor_id())) {
                 VERIFY_SCHED_STATE(TASK_SCHEDULED | SHOULD_SCHEDULE | TASK_PICKED | WILL_SCHEDULE);
-                if (is_in_sched_state(TASK_PICKED | PICKED_WRONG_TASK))
+                if (is_in_sched_state(TASK_PICKED | PICKED_WRONG_TASK)) {
+                        TRACE_TASK(tsk, "Wrong task\n");
                         set_sched_state(PICKED_WRONG_TASK);
-                else
+                } else
                         set_sched_state(WILL_SCHEDULE);
         } /* else */
           /*    /\* Litmus tasks should never be subject to a remote */
@@ -32,8 +33,8 @@ void sched_state_will_schedule(struct task_struct* tsk)
           /*    BUG_ON(is_realtime(tsk)); */
 
 #ifdef CONFIG_PREEMPT_STATE_TRACE
-        TRACE_TASK(tsk, "set_tsk_need_resched() ret:%p\n",
-                   __builtin_return_address(0));
+        STRACE("%d: set_tsk_need_resched() ret:%p\n",
+              tsk->pid, __builtin_return_address(0));
 #endif
 }
 
@@ -69,9 +70,11 @@ void litmus_reschedule(int cpu)
          * is not aware of the need to reschedule at this point. */
 
         /* is a context switch in progress? */
-        if (cpu_is_in_sched_state(cpu, TASK_PICKED))
+        if (cpu_is_in_sched_state(cpu, TASK_PICKED)) {
+                STRACE("Transition onto wrong task\n");
                 picked_transition_ok = sched_state_transition_on(
                         cpu, TASK_PICKED, PICKED_WRONG_TASK);
+        }
 
         if (!picked_transition_ok &&
             cpu_is_in_sched_state(cpu, TASK_SCHEDULED)) {
@@ -90,17 +93,18 @@ void litmus_reschedule(int cpu)
                         smp_send_reschedule(cpu);
         }
 
-        TRACE_STATE("%s picked-ok:%d sched-ok:%d\n",
-                    __FUNCTION__,
-                    picked_transition_ok,
-                    scheduled_transition_ok);
+        STRACE("%s picked-ok:%d sched-ok:%d\n",
+               __FUNCTION__,
+               picked_transition_ok,
+               scheduled_transition_ok);
 }
 
 void litmus_reschedule_local(void)
 {
-        if (is_in_sched_state(TASK_PICKED))
+        if (is_in_sched_state(TASK_PICKED)) {
+                STRACE("Rescheduling into wrong task\n");
                 set_sched_state(PICKED_WRONG_TASK);
-        else if (is_in_sched_state(TASK_SCHEDULED | SHOULD_SCHEDULE)) {
+        } else if (is_in_sched_state(TASK_SCHEDULED | SHOULD_SCHEDULE)) {
                 set_sched_state(WILL_SCHEDULE);
                 set_tsk_need_resched(current);
         }
@@ -111,7 +115,7 @@ void litmus_reschedule_local(void)
 void sched_state_plugin_check(void)
 {
         if (!is_in_sched_state(TASK_PICKED | PICKED_WRONG_TASK)) {
-                TRACE("!!!! plugin did not call sched_state_task_picked()!"
+                STRACE("!!!! plugin did not call sched_state_task_picked()!"
                       "Calling sched_state_task_picked() is mandatory---fix this.\n");
                 set_sched_state(TASK_PICKED);
         }
diff --git a/litmus/sched_mc.c b/litmus/sched_mc.c
index 6edf86935a29..c8e50d30a483 100644
--- a/litmus/sched_mc.c
+++ b/litmus/sched_mc.c
@@ -30,7 +30,7 @@
 #include <litmus/dgl.h>
 #include <litmus/color.h>
 #include <litmus/way_tracker.h>
-#warning "MUST ADD CHECK FOR MAX WAYS"
+
 struct mc_signal {
         int                     update:1;
         int                     preempt:1;
@@ -207,6 +207,7 @@ static int cpu_lower_prio(struct bheap_node *a, struct bheap_node *b)
 static int mc_preempt_needed(struct domain *dom, struct task_struct* curr)
 {
         struct task_struct *next = dom->peek_ready(dom);
+
         if (!next || !curr) {
                 return next && !curr;
         } else {
@@ -223,10 +224,13 @@ static int mc_preempt_needed(struct domain *dom, struct task_struct* curr)
 static void update_crit_position(struct crit_entry *ce)
 {
         struct bheap *heap;
+
         if (is_global(ce->domain)) {
                 heap = domain_data(ce->domain)->heap;
+
                 BUG_ON(!heap);
                 BUG_ON(!bheap_node_in_heap(ce->node));
+
                 bheap_delete(cpu_lower_prio, heap, ce->node);
                 bheap_insert(cpu_lower_prio, heap, ce->node);
         }
@@ -239,6 +243,7 @@ static void update_crit_position(struct crit_entry *ce)
 static void fix_crit_position(struct crit_entry *ce)
 {
         struct server *server = &ce->server;
+
         if (is_global(ce->domain) && server->in_transit) {
                 server_state_change(server, server->state, 0);
                 update_crit_position(ce);
@@ -368,7 +373,7 @@ static void link_task_to_crit(struct crit_entry *ce,
                         server_state_change(ce_server, SS_ACTIVE, 0);
                 }
 
-                TRACE_MC_TASK(ce->server.linked, "Unlinking\n");
+                /* TRACE_MC_TASK(ce->server.linked, "Unlinking\n"); */
 
                 stop_crit(ce);
                 tsk_rt(ce->server.linked)->server.parent = 0;
@@ -552,9 +557,7 @@ static struct task_struct* preempt_crit(struct domain *dom, struct crit_entry *c
 
         /* Per-domain preemption */
         link_task_to_crit(ce, task);
-        /* if (old && can_requeue(old)) { */
-        /*      dom->requeue(dom, old); */
-        /* } */
+
         update_crit_position(ce);
 
         /* Preempt actual execution if this is a running task.
@@ -574,6 +577,7 @@ static struct task_struct* preempt_crit(struct domain *dom, struct crit_entry *c
 
 /**
  * update_crit_levels() - Update criticality entries for the new cpu state.
+ * Disables criticality levels lower than @entry's currenly linked task.
  * This should be called after a new task has been linked to @entry.
  * The caller must hold the @entry->lock, but this method will release it.
  */
@@ -585,6 +589,8 @@ static void update_crit_levels(struct cpu_entry *entry)
         struct task_struct *readmit[NUM_CRIT_LEVELS];
         enum crit_level level = entry_level(entry);
 
+        STRACE("Updating crit levels for cpu %d\n", entry->cpu);
+
         /* Remove lower priority tasks from the entry */
         for (i = level + 1; i < NUM_CRIT_LEVELS; i++) {
                 ce = &entry->crit_entries[i];
@@ -609,8 +615,10 @@ static void update_crit_levels(struct cpu_entry *entry)
                                 link_task_to_crit(ce, NULL);
                         }
                         TRACE_CRIT_ENTRY(ce, "Removing lower crit\n");
-                        server_state_change(server, SS_REMOVED, 1);
-
+                        server_state_change(server, SS_REMOVED,
+                                            is_global(ce->domain)?1:0);
+                } else {
+                        TRACE_CRIT_ENTRY(ce, "Already removed!\n");
                 }
         }
         /* Need to unlock so we can access domains */
@@ -669,21 +677,26 @@ static void check_global_preempt(struct domain *dom)
         }
 }
 
-static void check_partitioned_preempt(struct domain *dom)
+static void check_partitioned_preempt(struct cpu_entry *entry,
+                                      struct domain *dom)
 {
-        struct cpu_entry *entry;
-        struct crit_entry *ce;
+        struct crit_entry *ce = domain_data(dom)->crit_entry;
 
-        ce = domain_data(dom)->crit_entry;
-        entry = crit_cpu(ce);
+        /* Cache next task */
+        dom->peek_ready(dom);
+
+        raw_spin_lock(&entry->lock);
 
         if (ce->server.state == SS_REMOVED ||
             !mc_preempt_needed(dom, ce->server.linked)) {
-                return;
+                goto out_unlock;
         }
 
         entry->signal.preempt = 1;
         litmus_reschedule(entry->cpu);
+
+ out_unlock:
+        raw_spin_unlock(&entry->lock);
 }
 
 /**
@@ -701,12 +714,7 @@ static void check_for_preempt(struct domain *dom)
                 ce = domain_data(dom)->crit_entry;
                 entry = crit_cpu(ce);
 
-                /* Cache next task */
-                dom->peek_ready(dom);
-
-                raw_spin_lock(&entry->lock);
-                check_partitioned_preempt(dom);
-                raw_spin_unlock(&entry->lock);
+                check_partitioned_preempt(entry, dom);
         }
 }
 
@@ -798,25 +806,24 @@ static void job_completion(struct task_struct *task, int forced)
 
         if (lt_before(get_user_release(task), litmus_clock()) ||
             (release_server && tsk_rt(task)->completed)){
-                TRACE_TASK(task, "Executable task going back to running\n");
+                TRACE_MC_TASK(task, "Executable task going back to running\n");
                 tsk_rt(task)->completed = 0;
         }
 
         if (release_server || forced) {
-                /* TODO: Level A does this independently and should not */
-                if (release_server && CRIT_LEVEL_A != tsk_mc_crit(task)) {
-                        prepare_for_next_period(task);
-                }
+                prepare_for_next_period(task);
 
-                TRACE_TASK(task, "Is released: %d, now: %llu, rel: %llu\n",
-                           is_released(task, litmus_clock()), litmus_clock(),
-                           get_release(task));
+                TRACE_MC_TASK(task, "Is released: %d, now: %llu, rel: %llu\n",
+                              is_released(task, litmus_clock()), litmus_clock(),
+                              get_release(task));
 
                 /* Requeue non-blocking tasks */
                 if (is_running(task)) {
                         job_arrival(task);
                 }
         } else if (is_ghost(task)) {
+                BUG_ON(tsk_rt(task)->linked_on == NO_CPU);
+
                 entry = &per_cpu(cpus, tsk_rt(task)->linked_on);
                 ce = &entry->crit_entries[tsk_mc_crit(task)];
 
@@ -847,24 +854,27 @@ static enum hrtimer_restart mc_ghost_exhausted(struct hrtimer *timer)
 #endif
         struct task_struct *tmp = NULL;
         struct cpu_entry *entry = crit_cpu(ce);
-        TRACE("Firing here at %llu\n", litmus_clock());
-        TRACE_CRIT_ENTRY(ce, "For this\n");
+        int resched = 0;
+
+        TRACE_CRIT_ENTRY(ce, "Firing here at %llu\n", litmus_clock());
 
         raw_spin_lock(&entry->lock);
 
-        if (is_ghost(ce->server.linked)) {
+        if (ce->server.linked && is_ghost(ce->server.linked)) {
                 update_server_time(ce->server.linked);
                 if (budget_exhausted(ce->server.linked)) {
                         tmp = ce->server.linked;
                 }
         } else {
-                litmus_reschedule(crit_cpu(ce)->cpu);
+                resched = 1;
         }
 
         raw_spin_unlock(&entry->lock);
 
         if (tmp)
                 job_completion(tmp, 1);
+        else if (resched)
+                litmus_reschedule(entry->cpu);
 
 #ifndef CONFIG_MERGE_TIMERS
         return HRTIMER_NORESTART;
@@ -891,8 +901,6 @@ static lt_t __ce_timer_function(struct ce_dom_data *ce_data)
             ce->server.linked == ce_data->should_schedule)
         {
                 old_link = ce->server.linked;
-                link_task_to_crit(ce, NULL);
-                mc_ce_job_completion(dom, old_link);
         }
         raw_spin_unlock(&crit_cpu(ce)->lock);
 
@@ -900,7 +908,7 @@ static lt_t __ce_timer_function(struct ce_dom_data *ce_data)
 
         /* Job completion will check for preemptions by means of calling job
          * arrival if the task is not blocked */
-        if (NULL != old_link) {
+        if (old_link) {
                 STRACE("old_link " TS " so will call job completion\n", TA(old_link));
                 raw_spin_unlock(dom->lock);
                 job_completion(old_link, 1);
@@ -993,8 +1001,10 @@ static void mc_task_new(struct task_struct *t, int on_rq, int running)
                 entry = &per_cpu(cpus, task_cpu(t));
         t->rt_param._domain = entry->crit_entries[level].domain;
 
+#ifdef CONFIG_SCHED_TASK_TRACE
         tsk_rt(t)->flush = 0;
         tsk_rt(t)->load  = 0;
+#endif
 
         /* Userspace and kernelspace view of task state may differ.
          * Model kernel state as a budget enforced container
@@ -1098,16 +1108,17 @@ static void mc_task_exit(struct task_struct *task)
                 color_sched_out_task(task);
         }
 
+        /* TODO: restore. This was geting triggered by race conditions even when
+         * no level-A task was executing */
+        if (CRIT_LEVEL_A == tsk_mc_crit(task))
+                mc_ce_task_exit_common(task);
+
         remove_from_all(task);
         if (tsk_rt(task)->scheduled_on != NO_CPU) {
                 per_cpu(cpus, tsk_rt(task)->scheduled_on).scheduled = NULL;
                 tsk_rt(task)->scheduled_on = NO_CPU;
         }
 
-        /* TODO: restore. This was geting triggered by race conditions even when
-         * no level-A task was executing */
-        /* if (CRIT_LEVEL_A == tsk_mc_crit(task)) */
-        /*      mc_ce_task_exit_common(task); */
 
         local_irq_restore(flags);
 }
@@ -1259,30 +1270,30 @@ static void process_update_signal(struct cpu_entry *entry)
                 stop_crit(ce);
                 server_state_change(crit_server, SS_BLOCKED, 0);
         }
-
-
 }
 
-static void process_signals(struct cpu_entry *entry)
+static void process_preempt_signal(struct cpu_entry *entry)
 {
+        int i;
         struct domain *dom;
         struct crit_entry *ce;
-        struct mc_signal signal;
-        struct task_struct *preempted;
+        struct task_struct *preempted = NULL;
+        struct server *server;
 
-        ce  = &entry->crit_entries[CRIT_LEVEL_B];
-        dom = ce->domain;
+        STRACE("Reading preempt signal\n");
 
-        /* Load signals */
-        raw_spin_lock(&entry->signal_lock);
-        signal = entry->signal;
-        clear_signal(&entry->signal);
-        raw_spin_unlock(&entry->signal_lock);
+        for (i = 0; i < NUM_CRIT_LEVELS; i++) {
+                ce = &entry->crit_entries[i];
+                dom = ce->domain;
+                server = &ce->server;
+                preempted = NULL;
 
-        if (signal.preempt) {
+                /* Swap locks. We cannot acquire a domain lock while
+                 * holding an entry lock or deadlocks will happen
+                 */
                 raw_spin_lock(dom->lock);
-                /* A higher-priority task may exist */
-                STRACE("Reading preempt signal\n");
+
+                /* Do domain stuff before grabbing CPU locks */
                 dom->peek_ready(dom);
 
                 raw_spin_lock(&entry->lock);
@@ -1308,26 +1319,106 @@ static void process_signals(struct cpu_entry *entry)
                                 raw_spin_unlock(dom->lock);
                         }
 
-                        raw_spin_lock(&entry->lock);
+                        break;
                 } else {
+                        raw_spin_unlock(&entry->lock);
                         raw_spin_unlock(dom->lock);
                 }
-        } else {
-                raw_spin_lock(&entry->lock);
+        }
+}
+
+static void process_signals(struct cpu_entry *entry)
+{
+        struct mc_signal signal;
+
+        /* Load signals */
+        raw_spin_lock(&entry->signal_lock);
+        signal = entry->signal;
+        clear_signal(&entry->signal);
+        raw_spin_unlock(&entry->signal_lock);
+
+        if (signal.preempt) {
+                process_preempt_signal(entry);
         }
 
+        raw_spin_lock(&entry->lock);
+
         if (signal.update) {
                 process_update_signal(entry);
         }
 }
 
+static void reschedule_if_signaled(struct cpu_entry *entry)
+{
+        struct mc_signal signal;
+
+        raw_spin_lock(&entry->signal_lock);
+        signal = entry->signal;
+        raw_spin_unlock(&entry->signal_lock);
+
+        if (signal.update || signal.preempt) {
+                litmus_reschedule_local();
+        }
+}
+
+static void pre_schedule(struct task_struct *prev)
+{
+        lt_t exec, start = litmus_clock();
+
+        /* Update userspace exec time */
+        if (prev && tsk_rt(prev)->last_exec_time) {
+                exec = start - tsk_rt(prev)->last_exec_time;
+                tsk_rt(prev)->user_job.exec_time += exec;
+        }
+
+        /* Flush task pages */
+        if (prev && tsk_mc_crit(prev) == CRIT_LEVEL_B &&
+            is_realtime(prev) && get_rt_job(prev) > 1 && lock_cache) {
+                color_sched_out_task(prev);
+
+#ifdef CONFIG_SCHED_TASK_TRACE
+                tsk_rt(prev)->load += litmus_clock() - start;
+#endif
+        }
+
+        TS_LVLA_SCHED_START;
+        TS_LVLB_SCHED_START;
+        TS_LVLC_SCHED_START;
+}
+
+static void post_schedule(struct task_struct *next)
+{
+        lt_t start;
+
+        switch (tsk_mc_crit(next)) {
+        case CRIT_LEVEL_A: TS_LVLA_SCHED_END(next); break;
+        case CRIT_LEVEL_B: TS_LVLB_SCHED_END(next); break;
+        case CRIT_LEVEL_C: TS_LVLC_SCHED_END(next); break;
+        }
+
+        /* Cache in task pages */
+        if (tsk_mc_crit(next) == CRIT_LEVEL_B && lock_cache &&
+            get_rt_job(next) > 1) {
+                start = litmus_clock();
+
+                color_sched_in_task(next);
+
+#ifdef CONFIG_SCHED_TASK_TRACE
+                BUG_ON(tsk_rt(next)->load);
+                tsk_rt(next)->load = litmus_clock() - start;
+#endif
+        }
+
+        tsk_rt(next)->last_exec_time = litmus_clock();
+}
+
 /**
  * mc_schedule() - Return next task which should be scheduled.
  */
 static struct task_struct* mc_schedule(struct task_struct* prev)
 {
-        lt_t start, exec;
-        int out_of_time, sleep, preempt, exists, blocks, global, lower, work;
+
+        int out_of_time, sleep, preempt, exists, blocks, global, lower;
         struct cpu_entry* entry = &__get_cpu_var(cpus);
         struct task_struct *next = NULL;
 
@@ -1340,22 +1431,7 @@ static struct task_struct* mc_schedule(struct task_struct* prev)
                 low_prio_arrival(entry->will_schedule);
         }
 
-        if (prev && tsk_rt(prev)->last_exec_time) {
-                exec = litmus_clock() - tsk_rt(prev)->last_exec_time;
-                tsk_rt(prev)->user_job.exec_time += exec;
-        }
-
-        if (prev && tsk_mc_crit(prev) == CRIT_LEVEL_B &&
-            is_realtime(prev) && get_rt_job(prev) > 1 && lock_cache) {
-                start = litmus_clock();
-                work = color_sched_out_task(prev);
-                tsk_rt(prev)->flush = litmus_clock() - start;
-                ++tsk_rt(prev)->flush_work;
-        }
-
-        TS_LVLA_SCHED_START;
-        TS_LVLB_SCHED_START;
-        TS_LVLC_SCHED_START;
+        pre_schedule(prev);
 
         raw_spin_lock(&entry->lock);
 
@@ -1406,9 +1482,7 @@ static struct task_struct* mc_schedule(struct task_struct* prev)
                         job_arrival(entry->scheduled);
         }
 
-        /* TODO: move this down somehow */
-        sched_state_task_picked();
-
+        /* Acquires the entry lock */
         process_signals(entry);
 
         /* Pick next task if none is linked */
@@ -1424,23 +1498,12 @@ static struct task_struct* mc_schedule(struct task_struct* prev)
 
         raw_spin_unlock(&entry->lock);
 
-        if (next) {
-                switch (tsk_mc_crit(next)) {
-                case CRIT_LEVEL_A: TS_LVLA_SCHED_END(next); break;
-                case CRIT_LEVEL_B: TS_LVLB_SCHED_END(next); break;
-                case CRIT_LEVEL_C: TS_LVLC_SCHED_END(next); break;
-                }
-        }
+        sched_state_task_picked();
 
-        if (next && tsk_mc_crit(next) == CRIT_LEVEL_B && lock_cache && get_rt_job(next) > 1) {
-                start = litmus_clock();
-                work = color_sched_in_task(next);
-                tsk_rt(next)->load = litmus_clock() - start;
-                tsk_rt(next)->load_work = work;
-        }
+        reschedule_if_signaled(entry);
 
         if (next) {
-                tsk_rt(next)->last_exec_time = litmus_clock();
+                post_schedule(next);
                 TRACE_MC_TASK(next, "Picked this task\n");
         } else {
                 STRACE("CPU %d idles at %llu\n", entry->cpu, litmus_clock());
diff --git a/litmus/sched_mc_ce.c b/litmus/sched_mc_ce.c
index c799738c0090..9f6929e55ab2 100644
--- a/litmus/sched_mc_ce.c
+++ b/litmus/sched_mc_ce.c
@@ -118,9 +118,9 @@ unsigned int mc_ce_get_expected_job(const int cpu, const int idx)
 static inline lt_t get_cycle_offset(const lt_t when, const lt_t cycle_time)
 {
         long long st = atomic64_read(&start_time);
-        //lt_t offset = (when - st) % cycle_time;
-        lt_t offset = 0;
-        TRACE("when: %llu  cycle_time: %llu start_time: %lld  offset %llu\n",
+        lt_t offset = (when - st) % cycle_time;
+        /* lt_t offset = 0; */
+        STRACE("when: %llu  cycle_time: %llu start_time: %lld  offset %llu\n",
                         when, cycle_time, st, offset);
         return offset;
 }
@@ -138,7 +138,7 @@ void mc_ce_job_completion(struct domain *dom, struct task_struct *ts)
         const struct ce_pid_entry *pid_entry = get_pid_entry(cpu, idx);
         unsigned int just_finished;
 
-        TRACE_TASK(ts, "Completed\n");
+        TRACE_MC_TASK(ts, "Completed\n");
 
         /* sched_trace_task_completion(ts, 0); */
         /* post-increment is important here */
@@ -152,13 +152,13 @@ void mc_ce_job_completion(struct domain *dom, struct task_struct *ts)
         if (just_finished < pid_entry->expected_job) {
                 /* this job is already released because it's running behind */
                 set_rt_flags(ts, RT_F_RUNNING);
-                TRACE_TASK(ts, "appears behind: the expected job is %u but "
+                TRACE_MC_TASK(ts, "appears behind: the expected job is %u but "
                                 "job %u just completed\n",
                                 pid_entry->expected_job, just_finished);
         } else if (pid_entry->expected_job < just_finished) {
-                printk(KERN_CRIT "job %u completed in expected job %u which "
-                                "seems too early\n", just_finished,
-                                pid_entry->expected_job);
+                TRACE_MC_TASK(ts, "job %u completed in expected job %u which "
+                              "seems too early\n", just_finished,
+                              pid_entry->expected_job);
         }
 }
 
@@ -189,7 +189,7 @@ static int mc_ce_schedule_at(const struct domain *dom, lt_t offset)
         }
         /* can only happen if cycle_time is not right */
         BUG_ON(pid_entry->acc_time > pid_table->cycle_time);
-        TRACE("schedule at returning task %d for CPU %d\n", idx, ce_data->cpu);
+        STRACE("schedule at returning task %d for CPU %d\n", idx, ce_data->cpu);
         return idx;
 }
 
@@ -211,7 +211,7 @@ static struct task_struct *mc_ce_schedule(struct task_struct *prev)
         exists = NULL != ce_data->scheduled;
         sleep = exists && RT_F_SLEEP == get_rt_flags(ce_data->scheduled);
 
-        TRACE("exists: %d, sleep: %d\n", exists, sleep);
+        STRACE("exists: %d, sleep: %d\n", exists, sleep);
 
         if (sleep)
                 mc_ce_job_completion(dom, ce_data->scheduled);
@@ -226,7 +226,7 @@ static struct task_struct *mc_ce_schedule(struct task_struct *prev)
         should_sched_asleep = should_sched_exists &&
                 RT_F_SLEEP == get_rt_flags(ce_data->should_schedule);
 
-        TRACE("should_sched_exists: %d, should_sched_blocked: %d, "
+        STRACE("should_sched_exists: %d, should_sched_blocked: %d, "
                         "should_sched_asleep: %d\n", should_sched_exists,
                         should_sched_blocked, should_sched_asleep);
 
@@ -248,7 +248,7 @@ static void mc_ce_finish_switch(struct task_struct *prev)
         struct domain *dom = get_domain_for(smp_processor_id());
         struct ce_dom_data *ce_data = dom->data;
 
-        TRACE("finish switch\n");
+        STRACE("finish switch\n");
 
         if (is_realtime(current) && CRIT_LEVEL_A == tsk_mc_crit(current))
                 ce_data->scheduled = current;
@@ -495,19 +495,19 @@ lt_t mc_ce_timer_callback_common(struct domain *dom)
                 }
         }
 
-        if (ce_data->should_schedule) {
-                get_deadline(should_schedule) =
-                        cycle_start_abs + pid_entry->acc_time;
-                get_release(should_schedule) = tsk_rt(should_schedule)->job_params.deadline -
-                        pid_entry->budget;
-                tsk_rt(should_schedule)->job_params.exec_time = 0;
-
-                TRACE_MC_TASK(should_schedule, "Released!\n");
-                set_rt_flags(ce_data->should_schedule, RT_F_RUNNING);
-                sched_trace_task_release(should_schedule);
-                sched_trace_server_release(-should_schedule->pid, get_rt_job(should_schedule),
-                                           tsk_rt(should_schedule)->job_params);
-        }
+        /* if (ce_data->should_schedule) { */
+        /*      get_deadline(should_schedule) = */
+        /*              cycle_start_abs + pid_entry->acc_time; */
+        /*      get_release(should_schedule) = tsk_rt(should_schedule)->job_params.deadline - */
+        /*              pid_entry->budget; */
+        /*      tsk_rt(should_schedule)->job_params.exec_time = 0; */
+
+        /*      TRACE_MC_TASK(should_schedule, "Released!\n"); */
+        /*      set_rt_flags(ce_data->should_schedule, RT_F_RUNNING); */
+        /*         sched_trace_task_release(should_schedule); */
+        /*         sched_trace_server_release(-should_schedule->pid, get_rt_job(should_schedule), */
+        /*                                 tsk_rt(should_schedule)->job_params); */
+        /* } */
         return next_timer_abs;
 }
 
@@ -535,7 +535,7 @@ static enum hrtimer_restart mc_ce_timer_callback(struct hrtimer *timer)
 #endif
         dom = get_domain_for(ce_data->cpu);
 
-        TRACE("timer callback on CPU %d (before lock)\n", ce_data->cpu);
+        STRACE("timer callback on CPU %d (before lock)\n", ce_data->cpu);
 
         raw_spin_lock_irqsave(dom->lock, flags);
         next_timer_abs = mc_ce_timer_callback_common(dom);
@@ -569,7 +569,7 @@ static int cancel_all_timers(void)
         int cancel_res;
 #endif
 
-        TRACE("cancel all timers\n");
+        STRACE("cancel all timers\n");
 
         for_each_online_cpu(cpu) {
                 dom = get_domain_for(cpu);
@@ -579,6 +579,8 @@ static int cancel_all_timers(void)
                 cancel_event(&ce_data->event);
 #else
                 cancel_res = hrtimer_cancel(&ce_data->timer);
+                atomic_set(&ce_data->timer_info.state,
+                           HRTIMER_START_ON_INACTIVE);
                 ret = ret || cancel_res;
 #endif
         }
@@ -598,7 +600,7 @@ static void arm_all_timers(void)
         int cpu, idx, cpu_for_timer;
         const lt_t start = atomic64_read(&start_time);
 
-        TRACE("arm all timers\n");
+        STRACE("arm all timers\n");
 
         for_each_online_cpu(cpu) {
                 dom = get_domain_for(cpu);
@@ -619,6 +621,9 @@ static void arm_all_timers(void)
                 add_event(get_event_group_for(cpu_for_timer),
                                 &ce_data->event, start);
 #else
+                hrtimer_start_on(cpu_for_timer, &ce_data->timer_info,
+                                 &ce_data->timer, ns_to_ktime(start),
+                                 HRTIMER_MODE_ABS_PINNED);
 #endif
         }
 }
@@ -630,7 +635,7 @@ static void arm_all_timers(void)
  */
 void mc_ce_release_at_common(struct task_struct *ts, lt_t start)
 {
-        TRACE("release CE at %llu\n", start);
+        STRACE("release CE at %llu\n", start);
         if (atomic_inc_and_test(&start_time_set)) {
                 /* in this case, we won the race */
                 cancel_all_timers();
diff --git a/litmus/sched_task_trace.c b/litmus/sched_task_trace.c
index 5a196e1b0fbd..c30fa1d74726 100644
--- a/litmus/sched_task_trace.c
+++ b/litmus/sched_task_trace.c
@@ -198,15 +198,9 @@ feather_callback void do_sched_trace_task_completion(unsigned long id,
         struct task_struct *t = (struct task_struct*) _task;
         struct st_event_record* rec = get_record(ST_COMPLETION, t);
         if (rec) {
-                rec->data.completion.exec   = tsk_rt(t)->user_job.exec_time;
-                rec->data.completion.flush  = tsk_rt(t)->flush;
-                rec->data.completion.load   = tsk_rt(t)->load;
-                rec->data.completion.flush_work = tsk_rt(t)->flush_work;
-                rec->data.completion.load_work  = tsk_rt(t)->load_work;
-                tsk_rt(t)->flush = 0;
+                rec->data.completion.when = now();
+                rec->data.completion.load = tsk_rt(t)->load;
                 tsk_rt(t)->load  = 0;
-                tsk_rt(t)->flush_work = 0;
-                tsk_rt(t)->load_work  = 0;
                 put_record(rec);
         }
 }
@@ -245,34 +239,6 @@ feather_callback void do_sched_trace_sys_release(unsigned long id,
         }
 }
 
-feather_callback void do_sched_trace_task_exit(unsigned long id,
-                                               unsigned long _task)
-{
-        struct task_struct *t = (struct task_struct*) _task;
-        const lt_t max_exec_time = tsk_rt(t)->max_exec_time;
-        const lt_t avg_exec_time = div64_u64(tsk_rt(t)->tot_exec_time, (get_job_no(t) - 1));
-
-        struct st_event_record *rec = get_record(ST_TASK_EXIT, t);
-        if (rec) {
-                rec->data.task_exit.avg_exec_time = avg_exec_time;
-                rec->data.task_exit.max_exec_time = max_exec_time;
-                put_record(rec);
-        }
-}
-
-feather_callback void do_sched_trace_task_tardy(unsigned long id,
-                                               unsigned long _task)
-{
-        struct task_struct *t = (struct task_struct*) _task;
-        struct st_event_record *rec = get_record(ST_TASK_TARDY, t);
-        if (rec) {
-                rec->data.task_tardy.max_tardy = tsk_rt(t)->max_tardy;
-                rec->data.task_tardy.total_tardy = tsk_rt(t)->total_tardy;
-                rec->data.task_tardy.missed = tsk_rt(t)->missed;
-                put_record(rec);
-        }
-}
-
 feather_callback void do_sched_trace_action(unsigned long id,
                                             unsigned long _task,
                                             unsigned long action)
diff --git a/litmus/way_tracker.c b/litmus/way_tracker.c
index ff392ab09c4d..9131c9658c2a 100644
--- a/litmus/way_tracker.c
+++ b/litmus/way_tracker.c
@@ -41,7 +41,9 @@ static int take_next_way(unsigned int color)
         } else {
                 printk(KERN_WARNING "Vury bad\n");
                 /* Seriously bad. */
+#ifdef CONFIG_KGDB
                 kgdb_breakpoint();
+#endif
                 BUG();
         }
         raw_spin_unlock(&lock);