Fix klmirqd plugin switching to not panic.

4 files changed, 35 insertions, 33 deletions

diff --git a/litmus/Kconfig b/litmus/Kconfig
index 9aeae659ae32..8ca66b4d687c 100644
--- a/litmus/Kconfig
+++ b/litmus/Kconfig
@@ -443,19 +443,6 @@ config NV_DEVICE_NUM
              Should be (<= to the number of CPUs) and
                  (<= to the number of GPUs) in your system.
 
-config NV_MAX_SIMULT_USERS
-        int "Maximum number of threads sharing a GPU simultanously"
-        depends on LITMUS_NVIDIA
-        range 1 3
-        default "2"
-        help
-                Should be at least equal to the #copy_engines + #execution_engines
-                of the GPUs in your system.
-
-                Scientific/Professional GPUs = 3  (ex. M2070, Quadro 6000?)
-                Consumer Fermi/Kepler GPUs   = 2  (GTX-4xx thru -6xx)
-                Older                        = 1  (ex. GTX-2xx)
-
 choice
           prompt "CUDA/Driver Version Support"
           default CUDA_5_0
diff --git a/litmus/litmus.c b/litmus/litmus.c
index fa244ba53e22..f98aa9d778a2 100644
--- a/litmus/litmus.c
+++ b/litmus/litmus.c
@@ -445,8 +445,8 @@ long __litmus_admit_task(struct task_struct* tsk)
 #endif
 #ifdef CONFIG_LITMUS_SOFTIRQD
         /* not an interrupt thread by default */
-        tsk_rt(tsk)->is_interrupt_thread = 0;
-    tsk_rt(tsk)->klmirqd_info = NULL;
+        //tsk_rt(tsk)->is_interrupt_thread = 0;
+    //tsk_rt(tsk)->klmirqd_info = NULL;
 #endif
 
         retval = litmus->admit_task(tsk);
@@ -523,11 +523,6 @@ static void synch_on_plugin_switch(void* info)
                 cpu_relax();
 }
 
-/* Switching a plugin in use is tricky.
- * We must watch out that no real-time tasks exists
- * (and that none is created in parallel) and that the plugin is not
- * currently in use on any processor (in theory).
- */
 int switch_sched_plugin(struct sched_plugin* plugin)
 {
         //unsigned long flags;
@@ -535,20 +530,21 @@ int switch_sched_plugin(struct sched_plugin* plugin)
 
         BUG_ON(!plugin);
 
+#ifdef CONFIG_LITMUS_SOFTIRQD
+        if (!klmirqd_is_dead()) {
+                kill_klmirqd();
+        }
+#endif
+
         /* forbid other cpus to use the plugin */
         atomic_set(&cannot_use_plugin, 1);
         /* send IPI to force other CPUs to synch with us */
         smp_call_function(synch_on_plugin_switch, NULL, 0);
 
         /* wait until all other CPUs have started synch */
-        while (atomic_read(&cannot_use_plugin) < num_online_cpus())
+        while (atomic_read(&cannot_use_plugin) < num_online_cpus()) {
                 cpu_relax();
-
-#ifdef CONFIG_LITMUS_SOFTIRQD
-        if (!klmirqd_is_dead()) {
-                kill_klmirqd();
         }
-#endif
 
         /* stop task transitions */
         //raw_spin_lock_irqsave(&task_transition_lock, flags);
@@ -571,6 +567,7 @@ int switch_sched_plugin(struct sched_plugin* plugin)
 out:
         //raw_spin_unlock_irqrestore(&task_transition_lock, flags);
         atomic_set(&cannot_use_plugin, 0);
+
         return ret;
 }
 
diff --git a/litmus/litmus_softirq.c b/litmus/litmus_softirq.c
index 44e2d38ad982..9c5ecab5e8d9 100644
--- a/litmus/litmus_softirq.c
+++ b/litmus/litmus_softirq.c
@@ -79,6 +79,7 @@ void kill_klmirqd(void)
         {
                 unsigned long flags;
                 struct list_head *pos;
+                struct list_head *q;
 
                 raw_spin_lock_irqsave(&klmirqd_state.lock, flags);
 
@@ -86,7 +87,7 @@ void kill_klmirqd(void)
 
                 klmirqd_state.shuttingdown = 1;
 
-                list_for_each(pos, &klmirqd_state.threads) {
+                list_for_each_safe(pos, q, &klmirqd_state.threads) {
                         struct klmirqd_info* info = list_entry(pos, struct klmirqd_info, klmirqd_reg);
 
                         if(info->terminating != 1)
@@ -96,7 +97,9 @@ void kill_klmirqd(void)
                                 flush_pending(info->klmirqd);
 
                                 /* signal termination */
+                                raw_spin_unlock_irqrestore(&klmirqd_state.lock, flags);
                                 kthread_stop(info->klmirqd);
+                                raw_spin_lock_irqsave(&klmirqd_state.lock, flags);
                         }
                 }
 
@@ -219,7 +222,7 @@ int launch_klmirqd_thread(int cpu, klmirqd_callback_t* cb)
 #define KLMIRQD_SLICE_NR_JIFFIES 1
 #define KLMIRQD_SLICE_NS ((NSEC_PER_SEC / HZ) * KLMIRQD_SLICE_NR_JIFFIES)
 
-static int set_litmus_daemon_sched(struct task_struct* tsk)
+static int become_litmus_daemon(struct task_struct* tsk)
 {
     int ret = 0;
 
@@ -249,6 +252,16 @@ static int set_litmus_daemon_sched(struct task_struct* tsk)
     return ret;
 }
 
+static int become_normal_daemon(struct task_struct* tsk)
+{
+        int ret = 0;
+
+        struct sched_param param = { .sched_priority = 0};
+        sched_setscheduler_nocheck(tsk, SCHED_NORMAL, &param);
+
+        return ret;
+}
+
 static int register_klmirqd(struct task_struct* tsk)
 {
         int retval = 0;
@@ -318,6 +331,7 @@ static int unregister_klmirqd(struct task_struct* tsk)
 
         /* remove the entry in the klmirqd thread list */
         list_del(&info->klmirqd_reg);
+        mb();
         --klmirqd_state.nr_threads;
 
         /* remove link to klmirqd info from thread */
@@ -687,7 +701,7 @@ static int run_klmirqd(void* callback)
         struct klmirqd_info* info = NULL;
         klmirqd_callback_t* cb = (klmirqd_callback_t*)(callback);
 
-        retval = set_litmus_daemon_sched(current);
+        retval = become_litmus_daemon(current);
     if (retval != 0) {
         TRACE_CUR("%s: Failed to transition to rt-task.\n", __FUNCTION__);
         goto failed;
@@ -696,7 +710,7 @@ static int run_klmirqd(void* callback)
         retval = register_klmirqd(current);
         if (retval != 0) {
                 TRACE_CUR("%s: Failed to become a klmirqd thread.\n", __FUNCTION__);
-                goto failed;
+                goto failed_sched_normal;
         }
 
         if (cb && cb->func) {
@@ -781,9 +795,10 @@ failed_unregister:
         /* remove our registration from klmirqd */
         unregister_klmirqd(current);
 
-failed:
-    litmus_exit_task(current);
+failed_sched_normal:
+        become_normal_daemon(current);
 
+failed:
         return retval;
 }
 
diff --git a/litmus/nvidia_info.c b/litmus/nvidia_info.c
index 7883296a7a18..3d38b168d9ba 100644
--- a/litmus/nvidia_info.c
+++ b/litmus/nvidia_info.c
@@ -322,7 +322,10 @@ int init_nvidia_info(void)
         else
         {
                 TRACE("%s : Could not find NVIDIA module!  Loaded?\n", __FUNCTION__);
-                return(-1);
+
+                init_nv_device_reg();
+                return(0);
+//              return(-1);
         }
 }