kern/llsync: improve concurrency

Keep a local copy of a processor registration state to avoid acquiring the
global lock when attempting to commit a checkpoint from an unregistered
processor.

3 files changed, 14 insertions, 12 deletions

diff --git a/kern/llsync.c b/kern/llsync.c
index d43bcefd..0fd973d0 100644
--- a/kern/llsync.c
+++ b/kern/llsync.c
@@ -51,7 +51,7 @@
 
 #define LLSYNC_NR_WORKS_WARN 10000
 
-struct llsync_cpu_checkpoint llsync_cpu_checkpoints[MAX_CPUS];
+struct llsync_cpu llsync_cpus[MAX_CPUS];
 
 /*
  * Global lock protecting the remaining module data.
@@ -208,6 +208,9 @@ llsync_register_cpu(unsigned int cpu)
         llsync_process_global_checkpoint(cpu);
 
     spinlock_unlock_intr_restore(&llsync_lock, flags);
+
+    assert(!llsync_cpus[cpu].registered);
+    llsync_cpus[cpu].registered = 1;
 }
 
 static void
@@ -232,7 +235,8 @@ llsync_unregister_cpu(unsigned int cpu)
 {
     unsigned long flags;
 
-    llsync_reset_checkpoint(cpu);
+    assert(llsync_cpus[cpu].registered);
+    llsync_cpus[cpu].registered = 0;
 
     spinlock_lock_intr_save(&llsync_lock, &flags);
 
@@ -256,14 +260,11 @@ llsync_commit_checkpoint(unsigned int cpu)
 {
     unsigned long flags;
 
-    if (!llsync_cpu_checkpoints[cpu].checked)
+    if (!(llsync_cpus[cpu].registered && llsync_cpus[cpu].checked))
         return;
 
     spinlock_lock_intr_save(&llsync_lock, &flags);
-
-    if (bitmap_test(llsync_registered_cpus, cpu))
-        llsync_commit_checkpoint_common(cpu);
-
+    llsync_commit_checkpoint_common(cpu);
     spinlock_unlock_intr_restore(&llsync_lock, flags);
 }
 
diff --git a/kern/llsync.h b/kern/llsync.h
index 273346fb..c58030d0 100644
--- a/kern/llsync.h
+++ b/kern/llsync.h
@@ -90,7 +90,7 @@ llsync_read_unlock(void)
 static inline void
 llsync_reset_checkpoint(unsigned int cpu)
 {
-    llsync_cpu_checkpoints[cpu].checked = 0;
+    llsync_cpus[cpu].checked = 0;
 }
 
 /*
@@ -101,7 +101,7 @@ llsync_reset_checkpoint(unsigned int cpu)
 static inline void
 llsync_checkin(unsigned int cpu)
 {
-    llsync_cpu_checkpoints[cpu].checked = 1;
+    llsync_cpus[cpu].checked = 1;
 }
 
 /*
diff --git a/kern/llsync_i.h b/kern/llsync_i.h
index 89a7fe83..034e368b 100644
--- a/kern/llsync_i.h
+++ b/kern/llsync_i.h
@@ -21,12 +21,13 @@
 #include <kern/param.h>
 
 /*
- * Per-processor flag indicating if a processor checked in.
+ * Per-processor data.
  */
-struct llsync_cpu_checkpoint {
+struct llsync_cpu {
+    int registered;
     int checked;
 } __aligned(CPU_L1_SIZE);
 
-extern struct llsync_cpu_checkpoint llsync_cpu_checkpoints[MAX_CPUS];
+extern struct llsync_cpu llsync_cpus[MAX_CPUS];
 
 #endif /* _KERN_LLSYNC_I_H */